Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Kwang‐Cheng ChenSuggested from co-authorship
- Alfred O. HeroSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works505 from public data
- ZOO2,281
An effective black-box attack that also only has access to the input (images) and the output (confidence scores) of a targeted DNN is proposed, sparing the need for training substitute models and avoiding the loss in attack transferability.
- Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!1,427
It is suggested that fine-tuning aligned LLMs introduces new safety risks that current safety infrastructures fall short of addressing -- even if a model's initial safety alignment is impeccable, it is not necessarily to be maintained after custom fine- Tuning.
- DBA: Distributed Backdoor Attacks against Federated Learning1,003
The distributed backdoor attack (DBA) is proposed --- a novel threat assessment framework developed by fully exploiting the distributed nature of FL that can evade two state-of-the-art robust FL algorithms against centralized backdoors.
- EAD: Elastic-Net Attacks to Deep Neural Networks via Adversarial Examples701
The authors' elastic-net attacks to DNNs (EAD) feature L1-oriented adversarial examples and include the state-of-the-art L2 attack as a special case, suggesting novel insights on leveraging L1 distortion in adversarial machine learning and security implications ofDNNs.
- Topology Attack and Defense for Graph Neural Networks: An Optimization Perspective552
A novel gradient-based attack method is presented that facilitates the difficulty of tackling discrete graph data and yields higher robustness against both different gradient based and greedy attack methods without sacrificing classification accuracy on original graph.
- Variational Quantum Circuits for Deep Reinforcement Learning483
This work reshapes classical deep reinforcement learning algorithms like experience replay and target network into a representation of variational quantum circuits, and uses a quantum information encoding scheme to reduce the number of model parameters compared to classical neural networks.
- AutoZOOM: Autoencoder-Based Zeroth Order Optimization Method for Attacking Black-Box Neural Networks460
Experimental results suggest that, by applying AutoZOOM to a state-of-the-art black-box attack (ZOO), a significant reduction in model queries can be achieved without sacrificing the attack success rate and the visual quality of the resulting adversarial examples.
- Adversarial T-Shirt! Evading Person Detectors in a Physical World448
This is the first work that models the effect of deformation for designing physical adversarial examples with respect to-rigid objects such as T-shirts and shows that the proposed method achieves74% and 57% attack success rates in the digital and physical worlds respectively against YOLOv2 and Faster R-CNN.
- 445
- Is Robustness the Cost of Accuracy? – A Comprehensive Study on the Robustness of 18 Deep Image Classification Models424
This paper thoroughly benchmark 18 ImageNet models using multiple robustness metrics, including the distortion, success rate and transferability of adversarial examples between 306 pairs of models, and reveals several new insights.
- Vision Transformers Are Robust Learners404
This work uses six different diverse ImageNet datasets concerning robust classification to conduct a comprehensive performance comparison of ViT models and SOTA convolutional neural networks (CNNs), Big-Transfer (Kolesnikov et al. 2020) and presents analyses that provide both quantitative and qualitative indications to explain why ViTs are indeed more robust learners.
- A Primer on Zeroth-Order Optimization in Signal Processing and Machine Learning: Principals, Recent Advances, and Applications356
This article provides a comprehensive review of Zeroth-order optimization, with an emphasis on showing the underlying intuition, optimization principles, and recent advances in convergence analysis.
- 316
- Ring-A-Bell! How Reliable are Concept Removal Methods for Diffusion Models?297
Ring-A-Bell is introduced, a model-agnostic red-teaming tool for T2I diffusion models that can be prepared in advance without prior knowledge of the target model, and can transform prompts that were originally regarded as safe to evade existing safety mechanisms, thus revealing the defects of the so-called safety mechanisms.
- Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examples269
This paper proposes a projected gradient method combined with group lasso and gradient regularization for sequence-to-sequence (seq2seq) models, whose inputs are discrete text strings and outputs have an almost infinite number of possibilities.
- 257
- Prompting4Debugging: Red-Teaming Text-to-Image Diffusion Models by Finding Problematic Prompts211
This work proposes Prompting4Debugging (P4D) as a debugging and red-teaming tool that automatically finds problematic prompts for diffusion models to test the reliability of a deployed safety mechanism, and demonstrates the efficacy of the P4D tool in uncovering new vulnerabilities of SD models with safety mechanisms.
- 206
- Reinforcement-Learning Based Portfolio Management with Augmented Asset Movement Prediction States202
Experiments on two real-world datasets validate the effectiveness of SARL over existing PM approaches, both in terms of accumulated profits and risk-adjusted profits and extensive simulations are conducted to demonstrate the importance of the proposed state augmentation.
- 196
- Decentralizing Feature Extraction with Quantum Convolutional Neural Network for Automatic Speech Recognition175
An in-depth study of different quantum circuit encoder architectures is conducted to provide insights into designing QCNN-based feature extractors and demonstrates a high correlation between the proposed QCNN features, class activation maps, and the input Mel-spectrogram.
- Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases172
A data-limited TrojanNet detector (TND) is proposed, which can detect a TrojanNet without accessing any data samples, and it is shown that such a TND can be built by leveraging the internal response of hidden neurons, which exhibits the Trojan behavior even at random noise inputs.
- Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark163
This study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance, and introduces novel enhancements to ZO optimization, including block-wise descent, hybrid training, and gradient sparsity.
- How to Backdoor Diffusion Models?153
This paper presents the first study on the robustness of diffusion models against backdoor attacks, and proposes BadDiffusion, a novel attack framework that engineers compromised diffusion processes during model training for backdoor implantation.
- 147
- CNN-Cert: An Efficient Framework for Certifying Robustness of Convolutional Neural Networks146
This work proposes a general and efficient framework, CNN-Cert, that is capable of certifying robustness on general convolutional neural networks and demonstrates by extensive experiments that this method outperforms state-of-the-art lower-bound-based certification algorithms in terms of both bound quality and speed.
- 144
- Principled Multilayer Network Embedding142
It is proved that comparing with regular link prediction methods, "layer co-analysis" achieved the best performance on most of the datasets, while "network aggregation" and "results aggregation" also have better performance than regular link Prediction methods.
- 140
- Attacking Visual Language Grounding with Adversarial Examples: A Case Study on Neural Image Captioning137
The proposed Show-and-Fool algorithm can successfully craft visually-similar adversarial examples with randomly targeted captions or keywords, and the adversarialExamples can be made highly transferable to other image captioning systems.
- CAT: Customized Adversarial Training for Improved Robustness132
This paper proposes a new algorithm, named Customized Adversarial Training (CAT), which adaptively customizes the perturbation level and the corresponding label for each training sample in adversarial training, and shows theoretically the CAT scheme improves the generalization.
- Understanding and Improving Visual Prompting: A Label-Mapping Perspective116
A new VP framework, termed ILM-VP (iterative label mapping-based visual prompting), which automatically re-maps the source labels to the target labels and progressively improves the target task accuracy of VP is proposed.
- Word Mover’s Embedding: From Word2Vec to Document Embedding115
The Word Mover’s Embedding (WME) is proposed, a novel approach to building an unsupervised document (sentence) embedding from pre-trained word embeddings that consistently matches or outperforms state-of-the-art techniques, with significantly higher accuracy on problems of short length.
- Sharp-MAML: Sharpness-Aware Model-Agnostic Meta Learning112
This work empirically demonstrate that Sharp-MAML and its computation-efficient variant can outperform the plain-vanilla MAML baseline and complement the empirical study with the convergence rate analysis and the generalization bound of Sharp-mAML.
- Optimizing molecules using efficient queries from property evaluations107
QMO is proposed, a generic query-based molecule optimization framework that exploits latent embeddings from a molecule autoencoder that improves the desired properties of an input molecule based on efficient queries, guided by a set of molecular property predictions and evaluation metrics.
- 102
- 102
- 94
- Model Reprogramming: Resource-Efficient Cross-Domain Machine Learning93
The methodology of model reprogramming is elucidates, existing use cases are summarized, a theoretical explanation of the success of the success of model reprogramming is provided, and a discussion on open-ended research questions and opportunities is discussed.
- A Theoretical Understanding of Shallow Vision Transformers: Learning, Generalization, and Sample Complexity93
It is shown that a proper token sparsification can improve the test performance by removing label-irrelevant and/or noisy tokens, including spurious correlations, and it is proved that a training process using stochastic gradient descent (SGD) leads to a sparse attention map.
- Towards Verifying Robustness of Neural Networks Against A Family of Semantic Perturbations93
The results show that the proposed Semantify-NN can support robustness verification against a wide range of semantic perturbations and an efficient refinement technique is proposed to further significantly improve the semantic certificate.
- PROVEN: Verifying Robustness of Neural Networks with a Probabilistic Approach92
Interestingly, PROVEN is derived from current state-of-the-art worst-case neural network robustness verifi-cation frameworks, and therefore it can provide probability certificates of neural network robustness with little computational overhead on top of existing methods such as Fast-Lin, CROWN and CNN-Cert.
- 88
- 86
- 85
- FLIP: A Provable Defense Framework for Backdoor Mitigation in Federated Learning83
This work theoretically analyze the connection among cross-entropy loss, attack success rate, and clean accuracy in this setting and proposes a trigger reverse engineering based defense and shows that the method can achieve robustness improvement with guarantee without affecting benign accuracy.
- SEAL: Safety-enhanced Aligned LLM Fine-tuning via Bilevel Data Selection82
This work proposes SEAL, a novel framework to enhance safety in LLM fine-tuning, which learns a data ranker based on the bilevel optimization to up rank the safe and high-quality fine-tuning data and down rank the unsafe or low-quality ones.
- Theoretical error performance analysis for variational quantum circuit based functional regression77
This work first put forth an end-to-end QNN, TTN-VQC, which consists of a quantum tensor network based on a tensor-train network (TTN) for dimensionality reduction and a VQC for functional regression, and characterize the optimization properties of TTn-VZC by leveraging the Polyak-Lojasiewicz condition.
- AI Explainability 360: An Extensible Toolkit for Understanding Data and Machine Learning Models77
The toolkit is an open-source Python toolkit featuring ten diverse and state-of-the-art explainability methods and two evaluation metrics, and an extensible software architecture is implemented that organizes methods according to their place in the AI modeling pipeline.
- 77
- QTN-VQC: an end-to-end learning framework for quantum neural networks67
The study provides theoretical insights into the quantum advantages of the end-to-end learning pipeline based on QTN-VQC from two perspectives, and focuses on using the QTN-VQC architecture to alleviate the Barren Plateau problem in the training stage.
- Curse or Redemption? How Data Heterogeneity Affects the Robustness of Federated Learning67
It is found data heterogeneity is more of a curse than a redemption as the attack effectiveness can be significantly boosted by simply adjusting the client-side backdooring timing, and data heterogeneity may result in overfitting at the local training of benign clients, which can be utilized by attackers to disguise themselves and fool skewed-feature based defenses.
- On the Design of Black-Box Adversarial Examples by Leveraging Gradient-Free Optimization and Operator Splitting Method64
This work introduces a general framework based on an operator splitting method, the alternating direction method of multipliers (ADMM) to devise efficient, robust black-box attacks that work with various distortion metrics and feedback settings without incurring high query complexity.
- Multilayer Spectral Graph Clustering via Convex Layer Aggregation: Theory and Algorithms63
This paper presents a multilayer spectral graph clustering (SGC) framework that performs convex layer aggregation and provides statistical clustering reliability guarantees and uses the phase transition criterion to proposed MIMOSA, which features automated cluster assignment and layer weight adaptation.
- 61
- Traffic-Aware Patching for Cyber Security in Mobile IoT61
A novel traffic-aware patching scheme to select important intermediate nodes to patch, which applies to the IoT system with limited patching resources and response time constraint, is proposed, which can effectively prevent the proliferation of malware propagation.
- On the Trustworthiness of Generative Foundation Models: Guideline, Assessment, and Perspective60
A holistic framework for advancing trustworthiness in GenAI is established, paving the way for safer and more responsible integration of GenFMs into critical applications and releasing the toolkit for dynamic evaluation.
- Larimar: Large Language Models with Episodic Memory Control60
Larimar is presented - a novel, brain-inspired architecture for enhancing LLMs with a distributed episodic memory that allows for dynamic, one-shot updates of knowledge without the need for computationally expensive re-training or fine-tuning.
- AI explainability 360: hands-on tutorial58
This tutorial will teach participants to use and contribute to a new open-source Python package named AI Explainability 360 (AIX360) (https://aix360.mybluemix.net), a comprehensive and extensible toolkit that supports interpretability and explainability of data and machine learning models.
- On the Convergence and Sample Complexity Analysis of Deep Q-Networks with $ε$-Greedy Exploration57
This paper provides the first theoretical convergence and sample complexity analysis of the practical setting of DQNs with $\epsilon$-greedy policy and proves an iterative procedure with decaying $\varepsilon$ converges to the optimal Q-value function geometrically.
- Characterizing Speech Adversarial Examples Using Self-Attention U-Net Enhancement57
A U-Net based attention model, UNetAt, is presented to enhance adversarial speech signals and it is found that temporal features learned by the attention network are capable of enhancing the robustness of DNN based ASR models.
- 56
- A deep fusion model for stock market prediction with news headlines and time series data52
This paper proposes a multimodal deep fusion model to predict stock trends, leveraging daily stock prices, technical indicators, and sentiment in daily news headlines published by media outlets, and demonstrates the positive impact of multimodal deep learning for stock trend prediction.
- Generating Contrastive Explanations with Monotonic Attribute Functions.52
This paper proposes a method that can generate contrastive explanations for deep neural networks where aspects that are in themselves sufficient to justify the classification by the deep model are highlighted, but also new aspects which if added will change the classification.
- RIGID: A Training-free and Model-Agnostic Framework for Robust AI-Generated Image Detection51
RIGID is a simple yet effective approach that identifies whether an image is AI-generated by comparing the representation similarity between the original and the noise-perturbed counterpart, and exhibits strong generalization across different image generation methods and robustness to image corruptions.
- 51
- Deep Community Detection51
This paper formulates the problem of detecting deep communities as multi-stage node removal that maximizes a new centrality measure, called the local Fiedler vector centrality (LFVC), at each stage and proves that a greedy node/edge removal strategy, based on successive maximization of LFVC, has bounded performance loss.
- Patch-level Routing in Mixture-of-Experts is Provably Sample-efficient for Convolutional Neural Networks50
Focusing on a supervised classification task using a mixture of two-layer convolutional neural networks (CNNs), this work shows for the first time that pMoE provably reduces the required number of training samples to achieve desirable generalization and outperforms its single-expert counterpart of the same or even larger capacity.
- Robust Mixture-of-Expert Training for Convolutional Neural Networks50
A new router-expert alternating Adversarial training framework for MoE is proposed, termed AdvMoE, which achieves 1% ~ 4% adversarial robustness improvement over the original dense CNN, and enjoys the efficiency merit of sparsity-gated MoE, leading to more than 50% inference cost reduction.
- Visual Prompting for Adversarial Robustness50
This work proposes a new VP method, termed Class-wise Adversarial Visual Prompting (C-AVP), to generate class-wise visual prompts so as to not only leverage the strengths of ensemble prompts but also optimize their interrelations to improve model robustness.
- 50
- Show-and-Fool: Crafting Adversarial Examples for Neural Image Captioning.50
This paper designs three approaches for crafting adversarial examples in image captioning: a targeted caption method; a targeted keyword method; and an untargeted method, and formulate the process of finding adversarial perturbations as optimization problems and design novel loss functions for efficient search.
- Protecting Neural Networks with Hierarchical Random Switching: Towards Better Robustness-Accuracy Trade-off for Stochastic Defenses48
This paper proposes hierarchical random switching (HRS), which protects neural networks through a novel randomization scheme, and proposes Defense Efficiency Score (DES), a comprehensive metric that measures the gain in unsuccessful attack attempts at the cost of drop in test accuracy of any defense.
- 47
- Scalable Spectral Clustering Using Random Binning Features47
This paper presents a novel scalable spectral clustering method using Random Binning features (RB) to simultaneously accelerate both similarity graph construction and the eigendecomposition and introduces a state-of-the-art SVD solver to effectively compute eigenvectors of a large sparse feature matrix generated by RB.
- Large Language Models are Efficient Learners of Noise-Robust Speech Recognition46
This work proposes to extract a language-space noise embedding from the N-best list to represent the noise conditions of source speech, which can promote the denoising process in GER and designs a knowledge distillation approach via mutual information estimation to distill the real noise information in audio embeddings to the authors' language embedding.
- Leveraging Latent Features for Local Explanations46
The new definition of "addition" uses latent features to move beyond the limitations of previous explanations and resolve an open question laid out in Dhurandhar, et.
- 46
- Stochastic inexact augmented Lagrangian method for nonconvex expectation constrained optimization45
This paper designs and analyzes stochastic inexact augmented Lagrangian methods (Stoc-iALM) to solve problems involving a nonconvex composite objective and nonconvex smooth functional constraints and establishes an oracle complexity result of O ( ε - 5) , better than the best-known O ( ε - 6) result.
- Robust Event Classification Using Imperfect Real-World PMU Data45
Numerical experiments using the real-world data set from the Western Interconnection of the U.S. power transmission grid show that the event classifiers trained under the proposed framework can achieve high classification accuracy while being robust against low-quality data.
- 44
- Towards Query-Efficient Black-Box Adversary with Zeroth-Order Natural Gradient Descent44
A ZO-NGD method is proposed to design the adversarial attacks, which incorporates the zeroth-order gradient estimation technique catering to the black-box attack scenario and the second-order natural gradient descent to achieve higher query efficiency.
- AI Explainability 360 Toolkit44
An open-source software toolkit featuring eight diverse state-of-the-art explainability methods, two evaluation metrics, and an extensible software architecture that organizes these methods according to their use in the AI modeling pipeline to improve transparency of machine learning models and provides a platform to integrate new explainability techniques as they are developed.
- When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers43
This paper theoretically prove the effectiveness of task addition in simultaneously learning a set of irrelevant or aligned tasks, as well as the success of task negation in unlearning one task from irrelevant or contradictory tasks, and proves the proper selection of linear coefficients for task arithmetic to achieve guaranteed generalization to out-of-domain tasks.
- Defensive Prompt Patch: A Robust and Interpretable Defense of LLMs against Jailbreak Attacks43
Defensive Prompt Patch is introduced, a novel prompt-based defense mechanism specifically designed to protect LLMs against such sophisticated jailbreak strategies, and not only outperforms existing defense strategies in balancing safety and functionality, but also provides a scalable and interpretable solution applicable to various LLM platforms.
- Rethinking normalization methods in federated learning43
This work uncovers external covariate shift problem in FL, which is caused by the independent local training processes on different devices, and shows that layer normalization is a better choice in FL which can mitigate the external covariates shift and improve the performance of the global model.
- When crowdsourcing meets mobile sensing: a social network perspective43
The structure of mobile sensing schemes is investigated and crowdsourcing methods for mobile sensing are introduced, Inspired by social networks, one can establish trust among participatory agents to leverage the wisdom of crowds forMobile sensing.
- 43
- Which Features are Learnt by Contrastive Learning? On the Role of Simplicity Bias in Class Collapse and Feature Suppression42
This work provides the first unified theoretically rigorous framework to determine which features are learnt by CL, and indicates that bias of (stochastic) gradient descent towards finding simpler solutions is a key factor in collapsing subclass representations and suppressing harder class-relevant features.
- LLM Agents Should Employ Security Principles41
This position paper argues that the well-established design principles in information security, commonly referred to as security principles, should be employed when deploying LLM agents at scale, and introduces AgentSandbox, a conceptual framework embedding these security principles to provide safeguards throughout an agent's life-cycle.
- Towards Certificated Model Robustness Against Weight Perturbations40
This work proposes an efficient approach to compute a certified robustness bound of weight perturbations, within which neural networks will not make erroneous outputs as desired by the adversary, and demonstrates significant improvements on the generalization ability of quantized networks through the robustness-aware quantization scheme.
- Diminution of context association memory structure in subjects with subjective cognitive decline40
It is proposed that context association and related memory structures are vulnerable in the SCD stage, and cortical thickness, subcortical gray matter volume, and fractional anisotropy are reduced.
- 40
- REFINE: Inversion-Free Backdoor Defense via Model Reprogramming38
REFINE is proposed, an inversion-free backdoor defense method based on model reprogramming that enhances the defense capabilities while maintaining model utility and its resistance to potential adaptive attacks.
- 38
- 37
- Phase Transitions in Spectral Community Detection36
A general model where two arbitrarily connected subnetworks are connected by random external edges is considered, showing that when one performs community detection via spectral clustering there exists an abrupt phase transition as a function of the random external edge connection probability.
- AutoVP: An Automated Visual Prompting Framework and Benchmark35
This work proposes AutoVP, an end-to-end expandable framework for automating VP design choices, along with 12 downstream image-classification tasks that can serve as a holistic VP-performance benchmark and shows that AutoVP outperforms the best-known current VP methods by a substantial margin.
- How Robust are Randomized Smoothing based Defenses to Data Poisoning?35
This work proposes a novel bilevel optimization based data poisoning attack that degrades the robustness guarantees of certifiably robust classifiers and highlights the importance of training-data quality in achieving high certified adversarial robustness.
- Multi-path routing with end-to-end statistical QoS Provisioning in Underlay Cognitive Radio Networks35
- Certified Robustness of Quantum Classifiers Against Adversarial Examples Through Quantum Noise34
The definition of differential privacy is connected and the quantum classifier trained with the natural presence of additive noise is differentially private and a certified robustness bound is derived to enable quantum classifiers to defend against adversarial examples supported by experimental results.
- Generalization Guarantee of Training Graph Convolutional Networks with Graph Topology Sampling34
This paper provides the first theoretical justification of graph topology sampling in training (up to) three-layer GCNs for semi-supervised node classification and tackles the nonconvex interaction of weights across layers, which is under-explored in the existing theoretical analyses of GCNs.
- 34
- Evading Real-Time Person Detectors by Adversarial T-shirt.34
This work proposes adversarial T-shirt, a robust physical adversarial example for evading person detectors even if it suffers from deformation due to a moving person's pose change, and extends the method to the ensemble attack setting against object detectors YOLOv2 and Faster R-CNN simultaneously.
- Multi-centrality graph spectral decompositions and their application to cyber intrusion detection34
This paper proposes a new PCA method for single graph analysis, called multi-centrality graph PCA (MC-GPCA), and a new dictionary learning method for ensembles of graphs, both based on spectral decomposition of multi- centrality matrices.
- Overload: Latency Attacks on Object Detection for Edge Devices33
The experimental results show that with latency attacks, the inference time of a single image can be increased ten times longer in reference to the normal setting, which presents a significant threat to systems with limited computing resources.
- An Overview of Trustworthy AI: Advances in IP Protection, Privacy-preserving Federated Learning, Security Verification, and GAI Safety Alignment33
This overview paper anchors on recent advances in four research hotspots of trustworthy AI with compelling and challenging security, privacy, and safety issues, and aims to inspire more research effort into guiding AI technologies towards beneficial purposes with greater robustness against malicious use intent.
- Network properties determine neural network performance32
Using network science and dynamical systems tools, the authors develop a framework for predicting the performance of artificial neural networks that enables a neural capacitance metric to universally capture a model’s generalization capability on a downstream task and predict model performance using only early training results.
- Unraveling the Connections between Privacy and Certified Robustness in Federated Learning Against Poisoning Attacks32
This paper investigates both user-level and instance- level privacy of DPFL and provides formal privacy analysis to achieve improved instance-level privacy and provides two robustness certification criteria: certified prediction and certified attack inefficacy for DPFL on both user and instance levels.
- RTL1/PEG11 imprinted in human and mouse brain mediates anxiety-like and social behaviors and regulates neuronal excitability in the locus coeruleus32
Findings demonstrate RTL1 is imprinted in brain, mediates emotional and social behaviors, and regulates excitability in LC neurons.
- Enhanced Adversarial Strategically-Timed Attacks Against Deep Reinforcement Learning32
This paper introduces timing-based adversarial strategies against a DRL-based navigation system by jamming in physical noise patterns on the selected time frames and proposes two adversarial agent models: one refers to online learning; another one is based on evolutionary learning.
- Towards A Unified Min-Max Framework for Adversarial Exploration and Robustness32
This work shows that attacking model ensembles, devising universal perturbation under multiple inputs or data transformations, and generalized AT over different types of attack models can be solved under a unified and theoretically principled min-max optimization framework.
- Where are We in Audio Deepfake Detection? A Systematic Analysis over Generative and Detection Models31
SONAR is introduced, a synthetic AI-Audio Detection Framework and Benchmark aiming at providing a comprehensive evaluation for distinguishing cutting-edge AI-synthesized auditory content, and reveals the generalization limitations of existing detection methods and demonstrates that foundation models exhibit stronger generalization capabilities.
- 31
- 31
- Benchmarking large language models on safety risks in scientific laboratories30
It is shown that current models remain far from meeting the reliability needed for safe laboratory operation, and the urgent need for specialized safety evaluation frameworks before deploying artificial intelligence systems in real laboratory settings is underscored.
- Microstructural differences in white matter tracts across middle to late adulthood: a diffusion MRI study on 7167 UK Biobank participants30
By analyzing the large volume of diffusion magnetic resonance imaging data available from the UK Biobank, the study has provided a detailed description of heterogeneous age-related differences in tract properties over the whole brain which generally supports the myelodegeneration hypothesis.
- 29
- AI Maintenance: A Robustness Perspective29
This proposal for AI maintenance facilitates robustness assessment, status tracking, risk scanning, model hardening, and regulation throughout the AI lifecycle, which is an essential milestone toward building sustainable and trustworthy AI ecosystems.
- A Word is Worth A Thousand Dollars: Adversarial Attack on Tweets Fools Stock Predictions29
This paper addresses the task of adversarial generation by solving combinatorial optimization problems with semantics and budget constraints with consistent success rates and cause significant monetary loss in trading simulation by simply concatenating a perturbed but semantically similar tweet.
- 29
- Neural-Brane: Neural Bayesian Personalized Ranking for Attributed Network Embedding29
Experimental results demonstrate the superiority of the proposed method over the state-of-the-art existing methods for network embedding, and evaluate the quality of vertex embedding produced by Neural-Brane by solving the node classification and clustering tasks on four real-world datasets.
- 29
- 29
- 28
- Genome Architecture Mediates Transcriptional Control of Human Myogenic Reprogramming28
Findings reveal a role for topological features of genome architecture in the initiation of transcriptional programs during TF-mediated human cellular reprogramming, which sets the stage for a critical transition observed at several genomic scales reflecting definitive adoption of the myogenic phenotype.
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts27
This paper theoretically proves that prioritizing the pruning of the experts with a smaller change of the routers l2 norm from the pretrained model guarantees the preservation of test accuracy, while significantly reducing the model size and the computational requirements.
- Holistic Adversarial Robustness of Deep Learning Models27
This paper provides a comprehensive overview of research topics and foundational principles of research methods for adversarial robustness of deep learning models, including attacks, defenses, verification, and novel applications.
- Discrete Attacks and Submodular Optimization with Applications to Text Classification.27
It is proved that this set function is submodular for some popular neural network text classifiers under simplifying assumption, and guarantees a $1-1/e$ approximation factor for attacks that use the greedy algorithm.
- AI Risk Management Should Incorporate Both Safety and Security26
It is advocated that stakeholders in AI risk management should be aware of the nuances, synergies, and interplay between safety and security, and unambiguously take into account the perspectives of both disciplines in order to devise mostly effective and holistic risk mitigation approaches.
- AI Explainability 360: Impact and Design26
The different ways in which users have experienced AI Explainability 360 have resulted in multiple types of impact and improvements in multiple metrics, highlighted by the adoption of the toolkit by the independent LF AI & Data Foundation.
- Validation of neuroimaging-based brain age gap as a mediator between modifiable risk factors and cognition26
Modality-specific brain age models were constructed and applied to a population-based cohort and confirmed that BAG can serve as a mediating indicator linking risk loadings to cognitive functioning, implicating its potential in the management of cognitive aging and dementia.
- Characteristic Examples: High-Robustness, Low-Transferability Fingerprinting of Neural Networks26
This paper proposes Characteristic Examples for effectively fingerprinting deep neural networks, featuring high-robustness to the base model against model pruning as well as low-transferability to unassociated models, taking both robustness and transferability into consideration.
- 26
- Learning Graph Topological Features via GAN26
This paper is the firstline research on combining the use of GANs and graph topological analysis and shows that the method produces subgraphs retaining a wide range of topological features, even in early reconstruction stages (unlike a single GAN, which cannot easily identify such features, let alone reconstruct the original graph).
- Node removal vulnerability of the largest component of a network26
It is proved that minimizing the largest component size is equivalent to solving a matrix one-norm minimization problem whose column vectors are orthogonal and sparse and they form a basis of the null space of the associated graph Laplacian matrix.
- Exploiting Low-Rank Tensor-Train Deep Neural Networks Based on Riemannian Gradient Descent With Illustrations of Speech Processing25
This work exploits a low-rank tensor-train deep neural network to build an end-to-end deep learning pipeline, namely LR-TT-DNN, and designs low-complexity hybrid tensor networks by considering trade-offs between the model complexity and practical performance.
- Convex Bounds on the Softmax Function with Applications to Robustness Verification25
Convex lower bounds and concave upper bounds on the softmax function are provided, which are compatible with convex optimization formulations for characterizing neural networks and other ML models.
- Low-Resource Music Genre Classification with Cross-Modal Neural Model Reprogramming25
Experimental results suggest that a neural model pre-trained on large-scale datasets can successfully perform music genre classification by using the new reprogramming method, Input-dependent NMR.
- What Improves the Generalization of Graph Transformers? A Theoretical Dive into the Self-attention and Positional Encoding24
It is demonstrated that self-attention and positional encoding enhance generalization by making the attention map sparse and promoting the core neighborhood during training, which explains the superior feature representation of Graph Transformers.
- Improving Privacy-Preserving Vertical Federated Learning by Efficient Communication with ADMM24
This paper introduces a VFL framework with multiple heads (VIM), which takes the separate contribution of each client into account, and enables an efficient decomposition of the VFL optimization objective to sub-objectives that can be iteratively tackled by the server and the clients on their own.
- Joint Edge-Model Sparse Learning is Provably Efficient for Graph Neural Networks24
This paper provides the first theoretical characterization of joint edge-model sparse learning from the perspective of sample complexity and convergence rate in achieving zero generalization error and proves analytically that both sampling important nodes and pruning neurons with the lowest-magnitude can reduce the sample complexityand improve convergence without compromising the test accuracy.
- Benchmarking machine learning robustness in Covid-19 genome sequence classification24
This paper introduces several ways to perturb SARS-CoV-2 genome sequences to mimic the error profiles of common sequencing platforms such as Illumina and PacBio and shows from experiments that some simulation-based approaches with different perturbation budgets are more robust (and accurate) than others for specific embedding methods to certain noise simulations on the input sequences.
- Comparing the effectiveness of board game-based and drill-based education programs in improving Taiwanese nurses' fire safety knowledge, attitudes, and behavior: A quasi-experimental study24
A board game-based fire education program is similar to a tabletop exercise, and drill-based programs more accurately reflect actual circumstances, and both methods can be applied based on the educational objectives and actual educational settings.
- 24
- Universal phase transition in community detectability under a stochastic block model24
It is shown that, as the community sizes become large, the asymptotic phase-transition threshold p* is equal to √[p1p2], where pi(i=1,2) is the within-community edge connection probability, which means the phase-Transition threshold is universal in the sense that it does not depend on the ratio of community sizes.
- Sequential defense against random and intentional attacks in complex networks24
The results show that the sequential defense mechanism greatly improves the network robustness and mitigates the possibility of network disruption by acquiring limited attack status information from a small subset of nodes in the network.
- Exploring the Benefits of Visual Prompting in Differential Privacy23
This work discovers that VP in tandem with PATE, a state-of-the-art DP training method that leverages the knowledge transfer from an ensemble of teachers, achieves the state-of-the-art privacy-utility tradeoff with minimum expenditure of privacy budget.
- Block Switching: A Stochastic Approach for Deep Learning Security23
Block Switching (BS) is introduced, a defense strategy against adversarial attacks based on stochasticity that replaces a block of model layers with multiple parallel channels, and the active channel is randomly assigned in the run time hence unpredictable to the adversary.
- Robust Text CAPTCHAs Using Adversarial Examples23
A user-friendly text-basedCAPTCHA generation method named Robust Text CAPTCHA (RTC), where the foregrounds and backgrounds are constructed with font and background images respectively sampled from font and image libraries, and they are synthesized into identifiable pseudo adversarial CAPTCHAs.
- 23
- Bias-Variance Tradeoff of Graph Laplacian Regularizer23
This letter presents a bias-variance tradeoff of graph Laplacian regularizer, which is widely used in graph signal processing and semisupervised learning tasks, and demonstrates near-optimal performance of the established analysis.
- Attack Atlas: A Practitioner's Perspective on Challenges and Pitfalls in Red Teaming GenAI22
The Attack Atlas is an intuitive framework that brings a practical approach to analyzing single-turn input attacks, placing it at the forefront for practitioners in the protection of generative AI systems.
- 22
- 22
- 22
- Causalainer: Causal Explainer for Automatic Video Summarization21
A Causal Explainer, dubbed Causalainer, is proposed to address the issue of explainability of video summarization by effectively distilling the mutual information from the visual and textual inputs in a causal semantics extractor.
- 21
- 21
- 21
- When Causal Intervention Meets Adversarial Examples and Image Masking for Deep Neural Networks21
Experimental results show that CE is a competitive and robust index for understanding DNNs when compared with conventional methods such as class-activation mappings on the Chest X-Ray-14 dataset for human-interpretable feature(s) (e.g., symptom) reasoning.
- On the Limitation of MagNet Defense Against L1-Based Adversarial Examples21
It is shown that adversarial examples crafted based on the L1 distortion metric can easily bypass MagNet and mislead the target DNN image classifiers on MNIST and CIFAR-10.
- FEAST: An Automated Feature Selection Framework for Compilation Tasks21
Intensive experimental results demonstrate that, under the compiler parameter assignment task, FEAST can achieve comparable results with about 18% of features that are automatically selected from the entire feature pool, hence reducing the accuracy and performance of the target compiler task.
- Diffusion Models for Information Dissemination Dynamics in Wireless Complex Communication Networks21
It is demonstrated that the proposed approach can generically characterize IDD in wireless complex networks and reveal salient features of dissemination dynamics in each network type, which could eventually aid in the design of more advanced, robust, and efficient networks and services.
- 20
- Linearity Grafting: Relaxed Neuron Pruning Helps Certifiable Robustness20
This work proposes a novel solution to strategically manipulate neurons, by "grafting" appropriate levels of linearity, to eliminate the non-linear components that are both redundant for DNN performance and harmful to its certification.
- Reinforcement learning based interconnection routing for adaptive traffic optimization20
The promise of applying reinforcement learning (RL) to optimize NoC runtime performance is demonstrated and three RL-based methods for learning optimal routing algorithms are presented.
- Accelerated Distributed Dual Averaging Over Evolving Networks of Growing Connectivity20
This paper extends the distributed dual averaging (DDA) subgradient algorithm to evolving networks of growing connectivity and provides quantitative evaluation of DDA acceleration for distributed optimization that is absent in the existing analysis.
- Analysis of Information Delivery Dynamics in Cognitive Sensor Networks Using Epidemic Models20
Simulation results show that the proposed interference-aware flooding schemes exploiting global timeout and vaccine recovery schemes to control the heavy buffer occupancy induced by packet replications can efficiently capture the complicated data delivery dynamics in CRAHNs in terms of end-to-end transmission reliability and buffer occupancy.
- Training a Resilient Q-network against Observational Interference19
This paper considers a deep q-network (DQN) framework training with an auxiliary task of observational interferences such as artificial noises and proposes a causal inference based DQN algorithm called causal inference Q- network (CIQ).
- AutoJailbreak: Exploring Jailbreak Attacks and Defenses through a Dependency Lens18
A systematic analysis of the dependency relationships in jailbreak attack and defense techniques, generalizing them to all possible attack surfaces and introducing a novel evaluation method that distinguishes hallucinations, which are often overlooked, from jailbreak attack and defense responses.
- A Quantum-Classical Collaborative Training Architecture Based on Quantum State Fidelity18
This article introduces a collaborative classical-quantum architecture called co-TenQu, which employs a tensor network for compression and feature extraction and proposes a quantum-state-fidelity-based evaluation function to iteratively train the network through a feedback loop between the two sides.
- Training Nonlinear Transformers for Chain-of-Thought Inference: A Theoretical Generalization Analysis17
This work provides the first theoretical study of training Transformers with nonlinear attention to obtain the CoT generalization capability so that the resulting model can inference on unseen tasks when the input is augmented by examples of the new task.
- 17
- 17
- 17
- 17
- Towards Mitigating Audio Adversarial Perturbations17
This work aims to explore the robustness of these audio adversarial examples generated via two attack strategies by applying different signal processing methods to recover the original audio sequence, and shows that by inspecting the temporal consistency in speech signals, it can potentially identify non-adaptive audio adversaries.
- 17
- 17
- MulBERRY: Enabling Bit-Error Robustness for Energy-Efficient Multi-Agent Autonomous Systems16
This work proposes MulBERRY, a multi-agent robust learning framework to enhance bit error robustness and energy efficiency for resource-constrained autonomous UAV swarms, and demonstrates the practicality of robust low-voltage operation on multi-UAV systems leading to energy savings in both compute and mission quality offlight.
- Detectors for Safe and Reliable LLMs: Implementations, Uses, and Limitations16
This work presents ongoing efforts to create and deploy a library of detectors: compact and easy-to-build classification models that provide labels for various harms.
- Distributed Adversarial Training to Robustify Deep Neural Networks at Scale16
DAT is general, which supports training over labeled and unlabeled data, multiple types of attack generation methods, and gradient compression operations favored for distributed optimization, and provides the convergence rate of DAT to the first-order stationary points in general non-convex settings.
- 16
- 16
- DiffuseKronA: A Parameter Efficient Fine-tuning Method for Personalized Diffusion Models15
Evaluated against diverse and complex input images and text prompts, DiffuseKronA consistently outperforms existing low-rank models, producing diverse images of higher quality with improved fidelity and a more accurate color distribution of objects, all the while upholding exceptional parameter efficiency.
- BERRY: Bit Error Robustness for Energy-Efficient Reinforcement Learning-Based Autonomous Systems15
BERRY supports robust learning, both offline and on-board the UAV, and for the first time, demonstrates the practicality of robust low-voltage operation on UAVs that leads to high energy savings in both compute-level operation and system-level quality-of-flight.
- 15
- 15
- 15
- A scalable attribute-aware network embedding system15
A scalable attribute-aware network embedding (SANE) method to efficiently learn the joint embedding representations from topology and attributes that provide more informative than a single representation extracted from topological structure or attributes alone is proposed.
- ON EXTENSIONS OF CLEVER: A NEURAL NETWORK ROBUSTNESS EVALUATION ALGORITHM15
A new formal robustness guarantee for classifier functions that are twice differentiable and the estimated robustness is called second-order CLEVER score is proposed.
- Phase Transitions and a Model Order Selection Criterion for Spectral Graph Clustering15
An automated model order selection (AMOS) is proposed, a solution to the SGC model selection problem under a random interconnection model using a novel selection criterion that is based on an asymptotic phase transition analysis.
- Revisiting Spectral Graph Clustering with Generative Community Models15
The presented method, SGC-GEN, not only considers the detection error caused by the corresponding model mismatch to a given graph, but also yields a theoretical guarantee on community detectability by analyzing Spectral Graph Clustering under GENerative community models (GCMs).
- Heterogeneous Aging Effects on Functional Connectivity in Different Cortical Regions: A Resting-State Functional MRI Study Using Functional Data Analysis15
Functional correlation analysis enables differentiation of changes in functional connectivities and characterizes the heterogeneous aging effects in different cortical regions.
- Action Recommendation for Cyber Resilience15
This paper presents an unifying graph-based model for representing the infrastructure, behavior and missions of an enterprise, and introduces an algorithm for recommending resilience establishing actions based on dynamic updates to the models.
- 15
- Understanding and Improving Training-Free AI-Generated Image Detections with Vision Foundation Models14
Experiments reveal that detection performance is closely linked to model robustness, with self-supervised (SSL) models providing more reliable representations and Contrastive Blur, which enhances performance on facial images, and MINDER, which addresses noise type bias, balancing performance across domains.
- 14
- Adversarial Examples Can Be Effective Data Augmentation for Unsupervised Machine Learning14
This paper proposes a framework of generating adversarial examples for unsupervised models and demonstrates novel applications to data augmentation, including data reconstruction, representation learning, and contrastive learning.
- Altered mismatch response of inferior parietal lobule in amnestic mild cognitive impairment: A magnetoencephalographic study14
The data suggest dysfunctional MMNm in patients with aMCI, particularly in the IPL, and the ROC curve analysis revealed that the combination of MMNn latencies of left IPL and CVVLT scores yielded a moderate accuracy in the discrimination of a MCI from HC at an individual level.
- 14
- 14
- Incremental eigenpair computation for graph Laplacian matrices: theory and applications14
It is demonstrated that users can utilize the proposed incremental method for effective eigenpair computation and for determining the desired number of clusters based on multiple clustering metrics.
- Decapitation via digital epidemics: a bio-inspired transmissive attack14
Transmissive attacks are described, the utility of epidemic models in communication systems is summarized, and connections between transmissive attacks and epidemic models are drawn.
- Optimal positions of clear aligner attachments to achieve lower canine tipping and bodily movement using finite element analysis13
The mesial occlusal–mesial cervical attachment configuration consistently had superior efficiency in facilitating both mesial and distal bodily movements of the canine and excelled in all four types of canine movement.
- Rethinking Backdoor Attacks on Dataset Distillation: A Kernel Method Perspective13
Following a comprehensive set of analyses and experiments, it is shown that the optimization-based trigger design framework informs effective backdoor attacks on dataset distillation, and datasets poisoned by the designed trigger prove resilient against conventional backdoor attack detection and mitigation methods.
- Be Your Own Neighborhood: Detecting Adversarial Example by the Neighborhood Relations Built on Self-Supervised Learning13
A novel AE detection framework, named BEYOND, for trustworthy predictions that can easily cooperate with the Adversarial Trained Classifier (ATC), achieving the state-of-the-art (SOTA) robustness accuracy.
- 13
- 13
- 13
- Quantum Machine Learning: An Interplay Between Quantum Computing and Machine Learning12
An overview of quantum computing for the machine learning paradigm, where variational quantum circuits (VQC) are used to develop QML architectures on noisy intermediate-scale quantum (NISQ) devices is presented.
- Learning on Transformers is Provable Low-Rank and Sparse: A One-layer Analysis12
This paper shows the first theoretical analysis of the property of low-rank and sparsity of one-layer Transformers by characterizing the trained model after convergence using stochastic gradient descent and concludes that proper magnitude-based pruning has a slight effect on the testing performance.
- MultiRobustBench: Benchmarking Robustness Against Multiple Attacks12
This paper presents the first unified framework for considering multiple attacks against ML models, able to model different levels of learner's knowledge about the test-time adversary, allowing it to model robustness against unforeseen attacks and robustnessagainst unions of attacks.
- Treatment Learning Causal Transformer for Noisy Image Classification12
This work proposes a transformer-based architecture, Treatment Learning Causal Transformer (TLT), that uses a latent generative model to estimate robust feature representations from current observational input for noise image classification and improves visual salience methods for perceiving noisy images.
- Causal Video Summarizer for Video Exploration12
A new causality-based method named Causal Video Summarizer (CVS) is proposed to effectively capture the interactive information between the video and query to tackle the task of multi-modal video summarization.
- Reprogramming Pretrained Language Models for Antibody Sequence Infilling12
Results on antibody design benchmarks show that the reprogrammed model on low resourced antibody sequence dataset provides highly diverse CDR sequences, up to more than a two-fold increase of diversity over the baselines, without losing structural integrity and naturalness.
- Learning to Design Fair and Private Voting Rules12
This paper introduces a novel notion of group fairness and adopt the existing notion of local differential privacy, and presents both a machine learning and a constrained optimization approach to design new voting rules that are fair while maintaining a high level of economic efficiency.
- TensorHyper-VQC: a tensor-train-guided hypernetwork for robust and scalable variational quantum computing11
TensorHyper-VQC is introduced, a novel tensor-train (TT)-guided hypernetwork framework that significantly improves the robustness and scalability of VQC and positions TensorHyper-VQC as a scalable and noise-resilient framework for advancing practical quantum machine learning on near-term devices.
- DDI-CoCo: A Dataset for Understanding the Effect of Color Contrast in Machine-Assisted Skin Disease Detection11
This work proposes a robust labeling method to quantify color contrast scores of each image and validate the method by showing small labeling variations, and observes a performance gap between the high and low color difference groups.
- 11
- 11
- Investigation of New Psychoactive Substances (NPS), Other Illicit Drugs, and Drug-Related Compounds in a Taiwanese Wastewater Sample Using High-Resolution Mass-Spectrometry-Based Targeted and Suspect Screening11
The HRMS-based comprehensive profile of NPS, other illicit drugs, and drug-related compounds in Taiwan is reported, which could be applied as biomarkers for estimating the consumption of drugs.
- Zeroth-Order Optimization for Composite Problems with Functional Constraints11
This appears to be the first work that develops an iALM-based ZO method for functional constrained optimization and meanwhile achieves query complexity results matching the best-known FO complexity results up to a factor of variable dimension.
- Self-Progressing Robust Training11
This paper proposes a new framework SPROUT, self-progressing robust training, and motivates SPROUT using a general formulation based on vicinity risk minimization, which includes many robust training methods as special cases.
- Rethinking Randomized Smoothing for Adversarial Robustness.11
This paper articulate and prove two major points: 1) the decision boundaries shrink with the adoption of randomized smoothing prediction rule; 2) noise augmentation does not necessarily resolve the shrinking issue and can even create additional issues.
- An Information-Theoretic Perspective on the Relationship Between Fairness and Accuracy.11
Fundamental limits on the relationship between fairness and accuracy are derived, which explain why the accuracy on a given dataset often decreases as fairness increases.
- 11
- 11
- PSBD: Prediction Shift Uncertainty Unlocks Backdoor Detection10
This paper proposes a novel method, Prediction Shift Backdoor Detection (PSBD), leveraging an uncertainty-based approach requiring minimal unlabeled clean validation data, and hypothesizes PS results from the neuron bias effect, making neurons favor features of certain classes.
- SF-DQN: Provable Knowledge Transfer using Successor Feature for Deep Reinforcement Learning10
The first convergence analysis with provable generalization guarantees for SF-DQN with GPI is established, revealing that SF-DQN with GPI outperforms conventional RL approaches in terms of both faster convergence rate and better generalization.
- CURE: A deep learning framework pre-trained on large-scale patient data for treatment effect estimation10
A new pre-training and fine-tuning framework, CURE (causal treatment effect estimation), for TEE from observational data, which is pre-trained on large-scale unlabeled patient data to learn representative contextual patient representations and fine-tuned on labeled patient data for TEE.
- The efficacy of a mindfulness-based exercise program in older residents of a long-term care facility in Taiwan10
This study evaluated the efficacy of a mindfulness-based exercise program in older residents of a long-term care facility in Taiwan and showed significantly larger improvements in a fear of falling, exercise self-efficacy, dynamic balance, and muscle strength in the experimental group than in the control group.
- 10
- Differential Associations of White Matter Brain Age With Language-Related Mechanisms in Word-Finding Ability Across the Adult Lifespan10
The results suggest that chronological aging and WM aging have differential effects on language-related word retrieval functions, and support that cognitive alterations in word-finding functions involve not only the domain-specific processing within the frontotemporal language network but also the Domain-general processing of executive functions in the fronto-parieto-occipital (or multi-demand) network.
- Towards an Efficient and General Framework of Robust Training for Graph Neural Networks10
On several applications, it is shown that the proposed techniques are significantly less computationally expensive and, in some cases, more robust than the state-of-the-art methods making them suitable to large-scale problems which were out of the reach of traditional robust training methods.
- Genome Architecture Leads a Bifurcation in Cell Identity10
It is found that chromatin reorganization occurred prior to significant transcriptional changes marking activation of the myogenic program, and a global bifurcation event delineated the transition into a myogenic cell identity 32 hours after exogenous MYOD1 activation.
- 10
- Conditional Modeling-Based Automatic Video Summarization9
A new approach to video summarization is proposed based on insights gained from how humans create ground truth video summaries that outperforms existing methods and achieves state-of-the-art performance on commonly used video summarizations datasets.
- Diversity Boosts AI-Generated Text Detection9
DivEye is a novel detection framework that captures how unpredictability fluctuates across a text using surprisal-based features, and provides interpretable insights into why a text is flagged, pointing to rhythmic unpredictability as a powerful and underexplored signal for LLM detection.
- 9
- Forecasting Open-Weight AI Model Growth on HuggingFace9
This work adapts the model introduced by Wang et al. for scientific citations, using three key parameters-immediacy, longevity, and relative fitness-to track the cumulative number of fine-tuned models of an open-weight model.
- 9
- 9
- Better May Not Be Fairer: A Study on Subgroup Discrepancy in Image Classification9
FlowAug is proposed, a semantic DA that leverages decoupled semantic representations captured by a pre-trained generative flow that achieves more consistent subgroup results than other types of DA methods on CIFAR10/100 and on CIFAR10/100-C.
- 9
- 9
- CARBEN: Composite Adversarial Robustness Benchmark9
This paper demonstrates how CAA's attack order affects the resulting image, and provides real-time inferences of different models, which will facilitate users' configuration of the parameters of the attack level and their rapid evaluation of model prediction.
- Neural Clamping: Joint Input Perturbation and Temperature Scaling for Neural Network Calibration9
This paper proposes a new post-processing calibration method called Neural Clamping, which employs a simple joint input-output transformation on a pre-trained classifier via a learnable universal input perturbation and an output temperature scaling parameter.
- Active Sampling of Multiple Sources for Sequential Estimation9
Owing to the sequential estimation being known to be analytically intractable, this paper adopts conditional estimation cost functions, leading to a sequential estimation approach that was recently shown to render tractable analysis.
- Non-Singular Adversarial Robustness of Neural Networks9
This study formalizes the notion of non-singular adversarial robustness for neural networks through the lens of joint perturbations to data inputs as well as model weights and demonstrates improved non- Singular robustness against joint input-weight adversarial perturbation.
- Fast Training of Provably Robust Neural Networks by SingleProp9
A new regularizer is developed that is both more efficient than existing certified defenses, requiring only one additional forward propagation through a network, and can be used to train networks with similar certified accuracy.
- 9
- 9
- Identifying Influential Links for Event Propagation on Twitter: A Network of Networks Approach9
Experimental results find that the between-network follower links are crucial to event propagation on Twitter, and utilizing the NoN model can successfully identify influential follower links such that their removals lead to a remarkable reduction in event propagate on Twitter follower networks.
- 9
- 8
- CENSOR: Defense Against Gradient Inversion via Orthogonal Subspace Bayesian Sampling8
This work presents a novel defense tailored for large neural network models that capitalizes on the high dimensionality of the model parameters to perturb gradients within a subspace orthogonal to the original gradient by leveraging cold posteriors over orthogonal subspaces.
- Combining Domain and Alignment Vectors to Achieve Better Knowledge-Safety Trade-offs in LLMs8
An efficient and effective merging-based alignment method called MergeAlign is introduced that interpolates the domain and alignment vectors, creating safer domain-specific models while preserving their utility.
- Large Language Models can be Strong Self-Detoxifiers8
Self-disciplined Autoregressive Sampling (SASA) is proposed, a lightweight controlled decoding algorithm for toxicity reduction of LLMs that markedly enhances the quality of the generated sentences relative to the original models and attains comparable performance to state-of-the-art detoxification techniques.
- 8
- Guest Editorial Special Issue on AI Enabled Cognitive Communication and Networking for IoT8
AI techniques, such as deep learning, probabilistic graph model, and reinforcement learning, aided with big data and IoT, provide a wide variety of tools and solutions to many new problems encountered in the design, operation, and optimization of cognitive communication systems and networking.
- IS ORDERED WEIGHTED ℓ1 REGULARIZED REGRESSION ROBUST TO ADVERSARIAL PERTURBATION? A CASE STUDY ON OSCAR8
Experimental results demonstrate that the regression performance of grouping strongly correlated features can be severely degraded under the authors' adversarial setting, even when the noise budget is significantly smaller than the ground-truth signals.
- 8
- Data-Driven Lipschitz Continuity: A Cost-Effective Approach to Improve Adversarial Robustness7
This work proposes a cost-efficient alternative based on Lipschitz continuity that achieves robustness comparable to models trained with extensive supplementary data and can integrate seamlessly with existing adversarial training frameworks and enhances the robustness of models without requiring extra generative data.
- ColorSense: A Study on Color Vision in Machine Visual Recognition7
The findings suggest that object recognition tasks such as classification and localization are susceptible to color vision bias, especially for high-stakes cases such as vehicle classes, and advanced mitigation techniques such as data augmentation and so on only give marginal improvement.
- The effect of a bone-preserving board game program on the knowledge, attitudes and preventive behaviors of osteoporosis in older adults7
The generalized estimating equation showed significantly larger improvements in knowledge, attitudes, and preventive behaviors in the experimental group than in the control group, providing a reference for educators, clinical practitioners and researchers.
- How Does Promoting the Minority Fraction Affect Generalization? A Theoretical Study of One-Hidden-Layer Neural Network on Group Imbalance7
It is shown that when all group-level co-variance is in the medium regime and all mean are close to zero, the learning performance is most desirable in the sense of a small sample complexity, a fast training rate, and a high average and group-level testing accuracy.
- Study on Anti-Inflammatory Effects of and Muscle Recovery Associated with Transdermal Delivery of Chaenomeles speciosa Extracts Using Supersonic Atomizer on Rat Model7
Pretreatment with CSEs yielded an immediate and significant increase in grip strength, with an increase of 180% and 165% occurring in the rats exposed to LC injection and exercise within the initial 12 h period, respectively, compared with the control group.
- The Devil is in the Neurons: Interpreting and Mitigating Social Biases in Pre-trained Language Models7
This work tries to unveil the mystery of social bias inside language models by introducing the concept of Social Bias Neurons and proposes Integrated Gap Gradients to accurately pinpoint units in a language model that can be attributed to undesirable behavior, such as social bias.
- 7
- Neural Model Reprogramming with Similarity Based Mapping for Low-Resource Spoken Command Recognition7
Experimental results show that with a pretrained AM trained on a large-scale English dataset, the proposed AR-SCR system outperforms the current state-of-the-art results on Arabic and Lithuanian speech commands datasets, with only a limited amount of training data.
- 7
- 7
- Fake it Till You Make it: Self-Supervised Semantic Shifts for Monolingual Word Embedding Tasks7
This work proposes a self-supervised approach to model lexical semantic change based on the perturbation of word vectors in the input corpora and shows that this method can be used for the detection of semantic change with any alignment method.
- Toward a neuro-inspired creative decoder7
It is revealed that atypical co-activation of highly activated and weakly activated neurons in a deep decoder promotes generation of novel and meaningful artifacts.
- Beyond Adversarial Training: Min-Max Optimization in Adversarial Attack and Defense.7
This work shows that attacking model ensembles, devising universal perturbation to input samples or data transformations, and generalized AT over multiple norm-ball threat models can be solved under a unified and theoretically principled min-max optimization framework.
- When Causal Intervention Meets Image Masking and Adversarial Perturbation for Deep Neural Networks7
Experimental results show that CE is a competitive and robust index for understanding DNNs when compared with conventional methods such as class-activation mappings on the ChestX-ray 14 dataset for human-interpretable feature(s) (e.g., symptom) reasoning.
- On the Supermodularity of Active Graph-Based Semi-Supervised Learning with Stieltjes Matrix Regularization7
This paper revisits the original formulation of graph-based SSL and proves the supermodularity of an AG-SSL objective function under a broad class of regularization functions parameterized by Stieltjes matrices, yielding a novel greedy label sampling algorithm with guaranteed performance relative to the optimal sampling set.
- Multilayer spectral graph clustering via convex layer aggregation7
A phase transition analysis is provided that establishes the existence of a critical value on the noise level that permits reliable cluster separation and specifies analytical upper and lower bounds on the critical value, where the bounds become exact when the clusters have identical sizes.
- 6
- 6
- Virus2Vec: Viral Sequence Classification Using Machine Learning6
Virus2Vec, a feature-vector representation for viral sequences that enable vector-space-based machine learning models to identify viral hosts, is proposed and empirically evaluated on real-world spike sequences of Coronaviridae and rabies virus sequence data to predict the host.
- 6
- Knowledge Graph Guided Simultaneous Forecasting and Network Learning for Multivariate Financial Time Series6
A framework GregNets is explored for jointly learning forecasting models and correlations structures that exploit graph connectivity from KGs and a pseudo-likelihood layer is developed that can learn the error residual structure for any multivariate time-series forecasting architecture in deep learning APIs (e.g. Tensorflow).
- 6
- 6
- 6
- 6
- TemPEST: Soft Template-Based Personalized EDM Subject Generation through Collaborative Summarization6
A novel personalized EDM subject generation model named Soft Template-based Personalized EDM Subject Generator (TemPEST) is proposed to consider the aforementioned users' characteristics when generating subjects, which contains a soft template-based selective encoder network, a user rating encodernetwork, a summary decoder network and a rating decoder.
- hpGAT: High-Order Proximity Informed Graph Attention Network6
A novel approach to appropriately define a variable receptive field for GNNs by incorporating high-order proximity information extracted from the hierarchical topological structure of the input graph by incorporating multiscale groups obtained from trainable hierarchical semi-nonnegative matrix factorization.
- 6
- Measuring the Robustness of Audio Deepfake Detection under Real-World Corruption5
This work systematically evaluates the robustness of 10 audio deepfake detection models against 18 common corruption types, grouped into categories: noise perturbation, audio modification, and compression, and yields four key insights.
- Optimization-Free Universal Watermark Forgery with Regenerative Diffusion Models5
This paper uncovers a greater risk of an optimization-free and universal watermark forgery that harnesses existing regenerative diffusion models and bypasses model retraining and enables adaptability to any image, presenting a greater challenge to the security of current watermarking techniques for diffusion models.
- Diagnostic spatio-temporal transformer with faithful encoding5
This paper addresses the task of anomaly diagnosis when the underlying data generation process has a complex spatio-temporal (ST) dependency, and formalizes the problem as supervised dependency discovery, where the ST dependency is learned as a side product of multivariate time-series classification.
- On Certifying and Improving Generalization to Unseen Domains5
It is demonstrated that the accuracy of the models trained with DG methods varies significantly across unseen domains, generated from popular benchmark datasets, and a universal certification framework based on distributionally robust optimization that can efficiently certify the worst-case performance of any DG method is proposed.
- Iterative Qubits Management for Quantum Index Searching in a Hybrid System5
IQuCS, which aims at index searching and counting in a quantum-classical hybrid system based on Grover’s algorithm, and reduces qubits consumption by up to 66.2%.
- 5
- 5
- 5
- 5
- 5
- 5
- Improving Prediction Efficacy Through Abnormality Detection and Data Preprocessing5
It is shown that corrupted data can be reconstructed through a Gaussian locally linear mappings method, and the prediction performance of adversarial samples can be improved by using the nearest neighbors as a surrogate.
- Dynamic Network Analysis of the 4D Nucleome5
A dynamic multilayer network approach to study the co-evolution of form and function in the 4D Nucleome as a temporal network with node dynamics, where the network topology is captured by chromosome conformation, and the function of a node is measured by RNA sequencing.
- 5
- 5
- Hey, That’s My Data! Token-Only Dataset Inference in Large Language Models4
CatShift is proposed, a token-only dataset inference framework based on catastrophic forgetting, where models overwrite prior knowledge when trained on new data, enabling practical protection of proprietary datasets.
- Getting Better at Working With You: Compiling User Corrections into Runtime Enforcement for Coding Agents4
Results suggest that compiling corrections into runtime enforcement can address a repeated-friction failure mode that memory alone does not reliably solve, reducing the need for users to restate the same correction across future sessions.
- 4
- 4
- Neural Network Reprogrammability: A Unified Theme on Model Reprogramming, Prompt Tuning, and Prompt Instruction4
This survey introduces neural network reprogrammability as a unifying framework that bridges mainstream model adaptation techniques--model reprogramming, prompt tuning, and prompt instruction--previously fragmented research areas yet converges on a shared principle: repurposing a pre-trained model by manipulating information at the interfaces while keeping the model parameters frozen.
- 4
- 4
- Lost In Translation: Generating Adversarial Examples Robust to Round-Trip Translation4
It is demonstrated that 6 state-of-the-art text-based adversarial attacks do not maintain their efficacy after round-trip translation, and an intervention-based solution is introduced by integrating Machine Translation into the process of adversarial example generation and demonstrating increased robustness to round- Trip translation.
- 4
- SynBench: Task-Agnostic Benchmarking of Pretrained Representations using Synthetic Data4
A new task-agnostic framework, SynBench, to measure the quality of pretrained representations using synthetic data and can be used to inform the design of robust linear probing on Pretrained representations to mitigate the robustness-accuracy tradeoff in downstream tasks.
- 4
- 4
- 4
- SChME at SemEval-2020 Task 1: A Model Ensemble for Detecting Lexical Semantic Change4
It is shown that languages that suffer less semantic change tend to benefit from using a large number of landmarks, whereas languages with more semantic change benefit from a more careful choice of landmark number for alignment.
- 4
- Recent Progress in Zeroth Order Optimization and Its Applications to Adversarial Robustness in Data Mining and Machine Learning4
One appealing application of ZO optimization to studying the robustness of deep neural networks - practical and efficient adversarial attacks that generate adversarial examples from a black-box machine learning model are highlighted.
- AMOS: An automated model order selection algorithm for spectral graph clustering4
This paper proposes AMOS, an automated model order selection algorithm for SGC that works by incrementally increasing the number of clusters, estimating the quality of identified clusters, and providing a series of clustering reliability tests.
- Phase transitions in spectral community detection of large noisy networks4
This paper proves phase transitions in community detectability as a function of the external edge connection probability and the noisy edge presence probability under a general network model where two arbitrarily connected communities are interconnected by random external edges.
- Supervised Collective Classification for Crowdsourcing4
The results on several crowdsourced data show that supervised methods can achieve better classification accuracy than unsupervised methods, and the proposed method outperforms other algorithms.
- 3
- 3
- The effect of the ‘Stable Sugar Winner’ board game on dietary knowledge, attitudes, and behavioral self-efficacy in preventing diabetes among community-dwelling older adults3
After completing intervention, the older adults had significantly higher mean scores for dietary knowledge, attitudes, and behavioral self-efficacy in preventing diabetes compared to before the intervention.
- Group-Adaptive Threshold Optimization for Robust AI-Generated Text Detection3
FairOPT, an algorithm for group-specific threshold optimization for probabilistic AI-text detectors, is introduced, which paves the way for more robust classification in AI-generated content detection via post-processing.
- Masking Improves Contrastive Self-Supervised Learning for ConvNets, and Saliency Tells You Where3
This work aims to alleviate the burden of including masking operation into the contrastive-learning framework for convolutional neural networks as an extra augmentation method by explicitly taking the saliency constraint into consideration in which the masked regions are more evenly distributed among the foreground and background for realizing the masking-based augmentation.
- Steal Now and Attack Later: Evaluating Robustness of Object Detection against Black-box Adversarial Attacks3
The feasibility of generating ghost objects in adversarial examples by extending the concept of "steal now, decrypt later" attacks is demonstrated, posing a significant threat to AI security.
- 3
- 3
- The role of genetic testing in adult patients with unexplained epilepsy3
In adult patients with unexplained epilepsy, genetic epilepsy is more associated with multisystem involvement and multiple structural anomalies but not family history of seizures or young-onset seizures.
- Leveraging Pre-Trained Neural Networks to Enhance Machine Learning with Variational Quantum Circuits3
- 3
- Inference and Denoise: Causal Inference-Based Neural Speech Enhancement3
Experimental evidence demonstrates that CISE outperforms a non-causal mask-based SE approach in the studied settings and has better performance and efficiency than more complex SE models.
- CURE: A Pre-training Framework on Large-scale Patient Data for Treatment Effect Estimation3
CURE is pre-trained on large-scale unlabeled patient data to learn representative contextual patient representations, and then fine-tuned on labeled patient data for TEE from observational data and incorporates structure and time into patient embeddings.
- SenSE: A Toolkit for Semantic Change Exploration via Word Embedding Alignment3
SenSE is presented, an interactive semantic shift exploration toolkit that provides visualization and explanation of lexical semantic change for an input pair of text sources as well as on explaining semantic change based on the neighbors of a chosen target word, while also extracting examples of sentences where these semantic deviations appear.
- 3
- 3
- 3
- AdvMS: A Multi-Source Multi-Cost Defense Against Adversarial Attacks3
A multi-source and multi-cost defense scheme that inherits advantages from two leading schemes: adversarial training and random model switching is proposed that mitigates the performance plateauing issue and enables improving robustness at a flexible and adjustable combination of costs over different factors which can better suit specific restrictions and needs in practice.
- 3
- HSCO-Bench: An Agent-Driven End-to-End Hardware-Software Co-design Benchmark for Systems-on-Chip2
HSCO-Bench is the first benchmark targeting this complete co-design flow, enabling LLMs to jointly reason about and modify both the software and hardware stacks of heterogeneous SoCs.
- Modular Prompt Learning Improves Vision-Language Models2
Modular Prompt Learning (MPL) is proposed that is designed to promote the preservation of information contained in the inserted prompts and is evaluated on base-to-new generalization and cross-dataset tasks.
- Computational Safety for Generative AI: A Hypothesis Testing Perspective2
A formalization of the concept of computational safety, which is a mathematical framework that enables the quantitative assessment, formulation, and study of safety challenges in GenAI through the lens of signal processing theory and methods, is presented.
- 2
- Model Reprogramming Demystified: A Neural Tangent Kernel Perspective2
It is demonstrated that the success of MR is governed by the eigenvalue spectrum of the NTK matrix on the target dataset and the critical role of the source model's effectiveness in determining reprogramming outcomes is established.
- 2
- ICX360: In-Context eXplainability 360 Toolkit2
In-Context Explainability 360 (ICX360), an open-source Python toolkit for explaining LLMs with a focus on the user-provided context (or prompts in general) that are fed to the LLMs, is introduced.
- Leveraging network motifs to improve artificial neural networks2
It is shown that incoherent network motifs provide greater stability and resilience to noise than coherent network motifs, offering new structural insights for designing stronger neural networks.
- 2
- CarBoN: Calibrated Best-of-N Sampling Improves Test-time Reasoning2
CarBoN (Calibrated Best-of-$N$), a two-phase method that first explores the solution space and then learns a calibration of the logits via an input-specific temperature $T$ and additive shift vector $\delta$, guiding generation toward more reliable reasoning.
- 2
- Graph is all you need? Lightweight data-agnostic neural architecture search without training2
This method, dubbed nasgraph, remarkably reduces the computational costs by converting neural architectures to graphs and using the average degree, a graph measure, as the proxy in lieu of the evaluation metric.
- MegaCOIN: Enhancing Medium-Grained Color Perception for Vision-Language Models2
It is shown that VLMs, including GPT-4o, have subpar color recognition capabilities, and fine-tuning with MegaCoin can result in improved performance on visual evaluation tasks.
- 2
- 2
- 2
- 2
- Evaluating the Adversarial Robustness for Fourier Neural Operators2
This research provides a sensitivity analysis tool and evaluation principles for assessing the adversarial robustness of ML-based scientific discovery models by generating adversarial examples for FNO, based on norm-bounded data input perturbations.
- 2
- 2
- 2
- Multimodal Brain Age Gap as a Mediating Indicator in the Relation between Modifiable Dementia Risk Factors and Cognitive Functioning2
A higher burden of modifiable dementia risk factors was causally associated with a greater cognitive decline, and this was significantly mediated by a larger multimodal BAG, which indicated an older brain.
- 2
- 2
- 2
- 2
- 2
- DEMO2
An algorithm for recommending resilience establishing actions based on dynamic updates to the models is introduced and its effectiveness will illustrate the effectiveness of the algorithm for preserving latency based quality of service (QoS).
- Considering Angle Selection When Using Ultrasound Electrode Displacement Elastography to Evaluate Radiofrequency Ablation of Tissues2
The experimental findings suggest that angle selection should be considered to achieve reliable electrode-displacement elastography to describe ablation zones.
- 2
- 1
- Defining the role of systemic autoimmune markers in adult epilepsy: A focus on autoimmune‐associated epilepsy1
This study investigates the utility of serum systemic autoantibodies as surrogate markers for positive neuronal antibodies and their potential in predicting epilepsy outcomes.
- 1
- 1
- 1
- 1
- 1
- When Does Visual Prompting Outperform Linear Probing for Vision-Language Models? A Likelihood Perspective1
A log-likelihood ratio (LLR) approach is proposed to analyze the comparative benefits of visual prompting and linear probing and attains up to a 100-fold reduction in run time compared to full training, while achieving prediction accuracies up to 91%.
- 1
- Patching LLM Like Software: A Lightweight Method for Improving Safety Policy in Large Language Models1
The results demonstrate that LLMs can be "patched" much like software, offering vendors and practitioners a practical mechanism for distributing scalable, efficient, and composable safety updates between major model releases.
- 1
- 1
- 1
- 1
- Model Reprogramming Outperforms Fine-tuning on Out-of-distribution Data in Text-Image Encoders1
The hidden costs associated with intrusive fine-tuning techniques are unveiled and a new model reprogramming approach for fine-tuning, which is name REPROGRAMMER, is introduced to improve the holistic performance of the downstream model across ID, OOD generalization, and OOD detection tasks.
- 1
- Improving Transformers using Faithful Positional Encoding1
This work proposes a new positional encoding method for a neural network architecture called the Transformer that is based on solid mathematical grounds and has a guarantee of not losing information about the positional order of the input sequence.
- Assessing the Resilience of Machine Learning Classification Algorithms on SARS-CoV-2 Genome Sequences Generated with Long-Read Specific Errors1
This study evaluates the accuracy and robustness of machine learning (ML) models using six different embedding techniques on SARS-CoV-2 error-incorporated genome sequences to gain valuable insights for researchers to effectively evaluate ML models and gain a better understanding of the approach for accurate identification of critical Sars-Cov-2 genome sequences.
- 1
- Understanding and improving zeroth-order optimization methods on AI-driven molecule optimization1
A parallel version of the Celada–Seiden cellular automaton that automates the very labor-intensive and therefore time-heavy and therefore expensive and expensive process of molecule optimization.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Third Workshop on Adversarial Learning Methods for Machine Learning and Data Mining (AdvML 2021)1
The AdvML'21 workshop consists of three tracks: (i) open-call paper submissions; (ii) invited speakers; and (iii) rising star awards and presentations.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Toward Cyber-Resiliency Metrics for Action Recommendations Against Lateral Movement Attacks1
The enterprise is modeled as a tripartite graph capturing the interactions between users, machines, and applications, and a set of procedures is proposed to harden the network by increasing the cost of lateral movement.
- 1
- 1
- ON THE UTILITY OF CONDITIONAL GENERATION BASED MUTUAL INFORMATION FOR CHARACTERIZING ADVERSARIAL SUBSPACES1
This paper uses MI as an information-theoretic metric to strengthen existing defenses and improve the performance of adversary detection, and demonstrates that the proposed MI detector can strengthen its robustness against powerful adversarial attacks.
- 1
- Ecology-Based DoS Attack in Cognitive Radio Networks1
A novel distributed denial-of-service (DoS) attack where invasive species, i.e., malicious users (MUs), induce originally normal-behaved SUs to execute concurrent transmissions with PUs and thus collapse the cognitive radio network is proposed.
- 1
- –
- Communication-Limited Multi-Agent Grid Congestion Management via Differentiable Optimization–
A novel framework for multi-agent grid congestion management inspired by the communication architecture used by RTE, the French transmission system operator, is presented and it is demonstrated that this approach significantly lowers cost compared to RTE’s existing baseline.
- –
- –
- –
- –
- –
- –
- –
- –
- Autoimmune and paraneoplastic encephalitis–associated neuronal autoantibodies with serial EEG in adults with unexplained new‐onset seizures or worsening epilepsy–
It is proposed that comprehensive antibody testing is beneficial for adults with new‐onset seizures or aggravated epilepsy without a predetermined etiology and a combination of serial electroencephalography (EEG) may enhance clinical management.
- –
- –
- –
- –
- –
- –
- –
- –
- SPARC: An AI-Based Speech Processing and Real-Time Correction System–
This project aims to develop a real-time voice correction system that automatically detects and corrects speech errors in near real-time while integrating the adjusted audio into ongoing conversations without disrupting the natural flow.
- Exploring the benefits of differentially private pre-training and fine-tuning for table transformers–
The benefits of combining differentially private pretraining, differentially private pretraining and fine-tuning of TabTransformers with a variety of Parameter-Efficient Fine-Tuning (PEFT) methods, including adapter, LoRA, and prompt tuning are explored.
- VP-NTK: Exploring the Benefits of Visual Prompting in Differentially Private Data Synthesis–
This work shows that VP in conjunction with DP-NTK, a DP generator that exploits the power of the neural tangent kernel (NTK) in training DP generative models, achieves a significant performance boost, particularly for high-resolution image datasets, with accuracy improving from 0.644±0.044 to 0.769.
- –
- –
- PEEL the Layers and Find Yourself: Revisiting Inference-Time Data Leakage for Residual Neural Networks–
A novel backward feature inversion method is proposed, PEEL, which can effectively recover block-wise input features from the intermediate output of residual NNs and outperforms the state-of-the-art recovery methods by an order of magnitude when evaluated by mean squared error (MSE).
- Abstract 2802: Circular RNA variants of theNUMB gene modulate the oncogenicity of oral squamous cell carcinoma–
This study concludes that circNUMBs mediate oncogenic functions opposite to linear NUMB transcript, and reveals that circNUMBs modulate cancer, cell development, cell injury, and cell movement genes.
- –
- –
- Linking Robustness and Generalization: A k* Distribution Analysis of Concept Clustering in Latent Space for Vision Models–
The k* Distribution, a local neighborhood analysis method, is used to examine the learned latent space at the level of individual concepts, which can be extended to examine the entire latent space.
- NaNa and MiGu: Semantic Data Augmentation Techniques to Enhance Protein Classification in Graph Neural Networks–
This paper proposes novel semantic data augmentation methods, Novel Augmentation of New Node Attributes (NaNa), and Molecular Interactions and Geometric Upgrading (MiGu) to incorporate backbone chemical and side-chain biophysical information into protein classification tasks and a co-embedding residual learning framework.
- –
- –
- –
- –
- –
- –
- Exploring the Benefits of Differentially Private Pre-training and Parameter-Efficient Fine-tuning for Table Transformers–
These extensive experiments show that these PEFT methods outperform traditional approaches in terms of the accuracy of the downstream task and the number of trainable parameters, thus achieving an improved trade-off among parameter efficiency, privacy, and accuracy.
- –
- –
- MENTOR: Multilingual Text Detection Toward Learning by Analogy–
“MENTOR” is proposed, the first work to realize a learning strategy between zero-shot learning and few- shot learning for multilingual scene text detection and can achieve comparable results with supervised methods for seen languages and outperform other methods in detecting unseen languages.
- –
- –
- Learning Geometrically Disentangled Representations of Protein Folding Simulations–
This work presents a geometric autoencoder framework to learn separate latent space encodings of the intrinsic and extrinsic geometries of the protein structure, and shows that this geometric learning-based method enjoys both accuracy and efficiency for generating complex structural variations.
- –
- –
- –
- –
- –
- The Fourth Workshop on Adversarial Learning Methods for Machine Learning and Data Mining (AdvML 2022)–
The AdvML'22 workshop consists of four tracks: (i) open-call paper submissions; (ii) invited speakers; (iii) rising star awards and presentations; and (iv) panel discussion on AdvML.
- –
- –
- –
- –
- –
- An Empirical Evaluation of Zeroth-Order Optimization Methods on AI-driven Molecule Optimization–
This paper studies the effectiveness of various ZO optimization methods for optimizing molecular objectives, which are characterized by variable smoothness, infrequent optima, and other challenges, and shows the advantages of ZO sign-based gradient descent.
- –
- –
- –
- –
- –
- –
- Editorial: Safe and Trustworthy Machine Learning–
A wide range of solutions are contributed as potentially viable solutions to address trust, safety and security issues faced by ML methods.
- –
- Domain Adaptation for Learning Generator From Paired Few-Shot Data–
This PFS-GAN aims to learn the disentangled representation from images, which composed of domain-invariant content features and domain-specific appearance features, which has better quantitative and qualitative results on the generated target-domain data with higher diversity in comparison to several baselines.
- –
- –
- –
- –
- –
- –
- –
- –
- Mild to moderate levels of anxiety in a healthy population cohort are related to face recognition with negative emotions and white matter integrity–
The face recognition tasks were applied to test the healthy population cohort with mild to moderate levels of anxiety and they underwent MRI scans to evaluate their brains' white matter integrity.
- –
- Corrections to “Learning Graph Topological Features via GAN”–
The authors have inadvertently left out three coauthors from the above paper - Hal Cooper, Min-Hwan Oh, and Sailung Yeung.
- SPROUT: Self-Progressing Robust Training–
This paper proposes a new framework SPROUT, self-progressing robust training, and motivates SPROUT using a general formulation based on vicinity risk minimization, which includes many robust training methods as special cases.
- –
- Optimizing Loss Landscape Connectivity via Neuron Alignment–
This work proposes a framework to investigate the effect of symmetry on the landscape connectivity by directly optimizing the weight permutations of the networks being connected by utilizing an existing neuron alignment technique and introduces a proximal alternating minimization scheme to address if an optimal permutation can be learned.
- Characterizing Adversarial Subspaces by Mutual Information–
Experimental results show that projected gradient descent, basic iterative method (BIM), Carlini and Wanger's attack (C&W attack) and elastic-net attack to deep neural network (elastic-net and L1 rules) can be effectively defended by the method designed.
- –
- Analysis and Actions on Graph Data.–
This dissertation focuses on the following topics in graph data analytics: Fundamental limits of spectral algorithms for graph clustering in single-layer and multilayer graphs, and efficient algorithms for actions on graphs, including graph spectral decompositions and insertions and removals of nodes or edges.
- –
- Sieve Estimators: Consistency and Rates of Convergence–
The approximation error is considered, which captures how well a class of hypotheses {Hk} approximates the Bayes decision boundary and considers the histogram classifiers in Figure 1 and Figure 2.
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.