Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks376 from public data
- LIBLINEAR: A Library for Large Linear Classification8,516
LIBLINEAR is an open source library for large-scale linear classification that supports logistic regression and linear support vector machines and provides easy-to-use command-line tools and library calls for users and developers.
- 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.
- Cluster-GCN1,634
Cluster-GCN is proposed, a novel GCN algorithm that is suitable for SGD-based training by exploiting the graph clustering structure and allows us to train much deeper GCN without much time and memory overhead, which leads to improved prediction accuracy.
- Can Decentralized Algorithms Outperform Centralized Algorithms? A Case Study for Decentralized Parallel Stochastic Gradient Descent1,564
This paper studies a D-PSGD algorithm and provides the first theoretical analysis that indicates a regime in which decentralized algorithms might outperform centralized algorithms for distributed stochastic gradient descent.
- 1,140
- A dual coordinate descent method for large-scale linear SVM1,043
A novel dual coordinate descent method for linear SVM with L1-and L2-loss functions that reaches an ε-accurate solution in O(log(1/ε)) iterations is presented.
- 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.
- Training and Testing Low-degree Polynomial Data Mappings via Linear SVM598
The proposed fast linear-SVM methods are applied to the explicit form of polynomially mapped data and successfully applied to a natural language processing (NLP) application by improving the testing accuracy under some training/testing speed requirements.
- ImageNet Training in Minutes473
This paper empirically evaluates the effectiveness on two neural networks: AlexNet and ResNet-50 trained with the ImageNet-1k dataset while preserving the state-of-the-art test accuracy, and uses large batch size, powered by the Layer-wise Adaptive Rate Scaling (LARS) algorithm, for efficient usage of massive computing resources.
- Towards Robust Neural Networks via Random Self-ensemble471
This paper proposes a new defense algorithm called Random Self-Ensemble (RSE), which adds random noise layers to the neural network to prevent the strong gradient-based attacks, and ensembles the prediction over random noises to stabilize the performance.
- 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.
- SSE-PT: Sequential Recommendation Via Personalized Transformer323
A Personalized Transformer (SSE-PT) model, outperforming SASRec by almost 5% in terms of NDCG@10 on 5 real-world datasets, and can handle extremely long sequences and outperform SASRec in ranking results with comparable training speed, striking a balance between performance and speed requirements.
- 316
- 306
- GenAttack305
GenAttack is introduced, a gradient-free optimization technique that uses genetic algorithms for synthesizing adversarial examples in the black-box setting and can successfully attack some state-of-the-art ImageNet defenses, including ensemble adversarial training and non-differentiable or randomized input transformations.
- 297
- 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.
- OR-Bench: An Over-Refusal Benchmark for Large Language Models268
This study proposes a novel method for automatically generating large-scale over-refusal datasets and introduces OR-Bench, the first large-scale over-refusal benchmark, to help the community develop better safety aligned models.
- Coordinate Descent Method for Large-scale L2-loss Linear Support Vector Machines267
A novel coordinate descent algorithm for training linear SVM with the L2-loss function that is more efficient and stable than state of the art methods such as Pegasos and TRON.
- A Comparison of Optimization Methods and Software for Large-scale L1-regularized Linear Classification262
Extensive comparisons indicate that carefully implemented coordinate descent methods are very suitable for training large document data.
- Scaling Up Dataset Distillation to ImageNet-1K with Constant Memory257
This work proposes a procedure to exactly compute the unrolled gradient with constant memory complexity, which allows to scale MTT to ImageNet-1K seamlessly with ~6x reduction in memory footprint, and discovers that it is challenging for M TT to handle datasets with a large number of classes, and proposes a novel soft label assignment that drastically improves its convergence.
- Fast coordinate descent methods with variable selection for non-negative matrix factorization256
This paper shows that FastHals has an inefficiency in that it uses a cyclic coordinate descent scheme and thus, performs unneeded descent steps on unimportant variables, and presents a variable selection scheme that uses the gradient of the objective function to arrive at a new coordinate descent method.
- 251
- QUIC: quadratic approximation for sparse inverse covariance estimation224
This work proposes a novel algorithm for solving the resulting optimization problem which is a regularized log-determinant program based on Newton's method and employs a quadratic approximation, but with some modifications that leverage the structure of the sparse Gaussian MLE problem.
- BIG & QUIC: Sparse Inverse Covariance Estimation for a Million Variables212
An algorithm BIGQUIC is developed, which can solve 1 million dimensional l1-regularized Gaussian MLE problems using a single machine, with bounded memory, and can achieve super-linear or even quadratic convergence rates.
- 210
- Self-Forcing++: Towards Minute-Scale High-Quality Video Generation201
This paper proposes a simple yet effective approach to mitigate quality degradation in long-horizon video generation without requiring supervision from long-video teachers or retraining on long video datasets, and substantially outperforms baseline methods in both fidelity and consistency.
- Towards Efficient and Scalable Sharpness-Aware Minimization183
A novel algorithm LookSAM is proposed - that only periodically calculates the inner gradient ascent, to significantly reduce the additional training cost of SAM and enjoys comparable computational complexity with first-order optimizers such as SGD or Adam.
- What Does BERT with Vision Look At?172
It is demonstrated that certain attention heads of a visually grounded language model actively ground elements of language to image regions, performing the task known as entity grounding.
- A sequential dual method for large scale multi-class linear svms167
Experiments indicate that the main idea is to sequentially traverse through the training set and optimize the dual variables associated with one example at a time, much faster than state of the art solvers such as bundle, cutting plane and exponentiated gradient methods.
- Low rank modeling of signed networks155
It is shown that the notion of weak structural balance in signed networks naturally leads to a global low-rank model for the network, which can be used for other analysis tasks on signed networks, such as user segmentation through signed graph clustering, with theoretical guarantees.
- 154
- Parallel matrix factorization for recommender systems153
It is shown that coordinate descent-based methods have a more efficient update rule compared to ALS and have faster and more stable convergence than SGD, and empirically show that CCD++ is much faster than ALS and SGD in both settings.
- Gradient boosted decision trees for high dimensional sparse output152
This paper studies the gradient boosted decision trees (GBDT) when the output space is high dimensional and sparse, and proposes a new GBDT variant, GBDT-SPARSE, to resolve this problem by employing L0 regularization.
- 147
- 145
- Nuclear Norm Minimization via Active Subspace Selection138
This is the first paper to scale nuclear norm solvers to the Yahoo-Music dataset, and the first time in the literature that the efficiency of nuclear normsolvers can be compared and even compete with non-convex solvers like Alternating Least Squares.
- 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.
- DrAttack: Prompt Decomposition and Reconstruction Makes Powerful LLM Jailbreakers136
It is discovered that decomposing a malicious prompt into separated sub-prompts can effectively obscure its underlying malicious intent by presenting it in a fragmented, less detectable form, thereby addressing these limitations.
- Matrix completion with noisy side information133
A novel model is proposed that balances between features and observations simultaneously in order to leverage feature information yet be robust to feature noise and it is shown that this model outperforms other methods for matrix completion that use features both in theory and practice.
- 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.
- FedDM: Iterative Distribution Matching for Communication-Efficient Federated Learning131
This work proposes FedDM to build the global training objective from multiple local surrogate functions, which enables the server to gain a more global view of the loss landscape and demonstrates that this method outperforms other FL counterparts in terms of efficiency and model performance given a limited number of communication rounds.
- NOMAD129
An efficient parallel distributed algorithm for matrix completion, named NOMAD (Non-locking, stOchastic Multi-machine algorithm for Asynchronous and Decentralized matrix completion), which outperforms synchronous algorithms which require explicit bulk synchronization after every iteration.
- Reducing BERT Pre-Training Time from 3 Days to 76 Minutes127
The LAMB optimizer is proposed, which helps to scale the batch size to 65536 without losing accuracy, and is a general optimizer that works for both small and large batch sizes and does not need hyper-parameter tuning besides the learning rate.
- 127
- 124
- 123
- ML-LOO: Detecting Adversarial Examples with Feature Attribution118
This work introduces a new framework to detect adversarial examples through thresholding a scale estimate of feature attribution scores, and extends it to include multi-layer feature attributions in order to tackle attacks that have mixed confidence levels.
- 115
- On the Robustness of Self-Attentive Models111
This work investigates the attention and feature extraction mechanisms of state-of-the-art recurrent neural networks and self-attentive architectures for sentiment analysis, entailment and machine translation under adversarial attacks and proposes a novel attack algorithm for generating more natural adversarial examples.
- 109
- A Review of Adversarial Attack and Defense for Classification Methods106
This article aims to introduce this topic and its latest developments to the statistical community, primarily focusing on the generation and guarding of adversarial examples, and to encourage more statisticians to work on this important and exciting field.
- 104
- 101
- Large-batch training for LSTM and beyond101
The Dynamic Adaptive-Tuning Engine (DATE) is proposed for better large-batch training and achieves a 5.3x average speedup over the baselines for four LSTM-based applications on the same hardware.
- Defense against Synonym Substitution-based Adversarial Attacks via Dirichlet Neighborhood Ensemble99
Dirichlet Neighborhood Ensemble is proposed, a randomized method for training a robust model to defense synonym substitution-based attacks that consistently outperforms recently proposed defense methods by a significant margin across different network architectures and multiple data sets.
- Emotional EEG classification using connectivity features and convolutional neural networks95
A new classification system that utilizes brain connectivity with a CNN is introduced and its effectiveness via the emotional video classification is validated by using three different types of connectivity measures.
- Searching for an Effective Defender: Benchmarking Defense against Adversarial Word Substitution94
An effective method to further improve the robustness of neural text classifiers against adversarial word-substitution attacks for neural NLP models is proposed and achieved the highest accuracy on both clean and adversarial examples on AGNEWS and IMDB datasets.
- 93
- 92
- 88
- 85
- 85
- Evaluating Robustness of Deep Image Super-Resolution Against Adversarial Attacks84
This paper investigates the robustness of deep learning-based super-resolution methods against adversarial attacks, which can significantly deteriorate the super-resolved images without noticeable distortion in the attacked low-resolution images.
- Robust and Accurate Object Detection via Adversarial Learning83
This work augments the fine-tuning stage for object detectors by exploring adversarial examples, which can be viewed as a model-dependent data augmentation, and dynamically selects the stronger adversarial images sourced from a detector’s classification and localization branches and evolves with the detector to ensure the augmentation policy stays current and relevant.
- Red Teaming Language Model Detectors with Language Models80
Experiments reveal that attacks effectively compromise the performance of all detectors in the study with plausible generations, underscoring the urgent need to improve the robustness of LLM-generated text detection systems.
- 79
- Watermarking Pre-trained Language Models with Backdooring78
It is shown that PLMs can be watermarked with a multi-task learning framework by embedding backdoors triggered by specific inputs defined by the owners, and those watermarks are hard to remove even though the watermarked PLMs are fine-tuned on multiple downstream tasks.
- 78
- RecurJac: An Efficient Recursive Algorithm for Bounding Jacobian Matrix of Neural Networks and Its Applications73
A recursive algorithm is proposed, RecurJac, to compute both upper and lower bounds for each element in the Jacobian matrix of a neural network with respect to network's input, and the network can contain a wide range of activation functions.
- 72
- A Scalable Asynchronous Distributed Algorithm for Topic Modeling70
It is shown that F+Nomad LDA significantly outperforms recent state-of-the-art topic modeling approaches on massive problems which involve millions of documents, billions of words, and thousands of topics.
- Large Linear Classification When Data Cannot Fit in Memory69
This work proposes and analyzes a block minimization framework for data larger than the memory size, and investigates two implementations of the proposed framework for primal and dual SVMs, respectively.
- 65
- 63
- Fast Prediction for Large-Scale Kernel Machines63
A new theoretical analysis on bounding the error of the solution computed by using Nystrom kernel approximation method, and it is shown that the error is related to the weighted kmeans objective function where the weights are given by the model computed from the original kernel.
- 100-epoch ImageNet Training with AlexNet in 24 Minutes.61
This work focuses on using large-batch data-parallelism synchronous SGD without losing accuracy in the fixed epochs for large-scale DNN training, and when the batch size is increased, its accuracy is much higher than Facebook's on corresponding batch sizes.
- RandomRooms: Unsupervised Pre-training from Synthetic Shapes and Randomized Layouts for 3D Object Detection60
A new method called RandomRooms is proposed to generate random layouts of a scene by making use of the objects in the synthetic CAD dataset and learn the 3D scene representation by applying object-level contrastive learning on two random scenes generated from the same set of synthetic objects.
- 60
- Lyapunov-stable Neural Control for State and Output Feedback: A Novel Formulation59
A novel formulation is proposed that defines a larger verifiable region-of-attraction (ROA) than shown in the literature, and refines the conventional restrictive constraints on Lyapunov derivatives to focus only on certifiable ROAs.
- Incorporating physics into data-driven computer vision58
Technical approaches for integrating physical models into artificial intelligence pipelines, referred to as physics-based machine learning, are discussed, which range from modifications to the dataset, network design, loss functions, optimization and regularization schemes.
- 57
- 55
- 54
- 52
- 52
- A Divide-and-Conquer Method for Sparse Inverse Covariance Estimation52
This paper considers the regularized sparse inverse covariance matrix estimation problem with a very large number of variables, and proposes a clustering algorithm that attempts to minimize this bound, and in practice, is able to find effective partitions of the variables.
- Evaluating and Enhancing the Robustness of Dialogue Systems: A Case Study on a Negotiation Agent51
Algorithm to evaluate the robustness of a dialogue agent by carefully designed attacks using adversarial agents performed in both black-box and white-box settings is developed and it is demonstrated that adversarial training using the authors' attacks can significantly improve the robustity of a goal-oriented dialogue system.
- Goal-Directed Inductive Matrix Completion51
Goal-directed Inductive Matrix Completion is proposed to learn a nonlinear mapping of the features so that they satisfy the required properties and the superiority of this method is established on several popular machine learning applications including multi-label learning, multi-class classification, and semi-supervised clustering.
- 50
- 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.
- 49
- Evaluating and Enhancing the Robustness of Neural Network-based Dependency Parsing Models with Adversarial Examples48
This study shows that adversarial examples also exist in dependency parsing, and proposes two approaches to study where and how parsers make mistakes by searching over perturbations to existing texts at sentence and phrase levels, and design algorithms to construct such examples in both of the black-box and white-box settings.
- 47
- Efficient Neural Interaction Function Search for Collaborative Filtering46
Experimental results demonstrate that the proposed method can be much more efficient than popular AutoML approaches, can obtain much better prediction performance than state-of-the-art CF approaches, and can discover distinct IFCs for different data sets and tasks.
- Iterative Scaling and Coordinate Descent Methods for Maximum Entropy Models45
A general and unified framework for iterative scaling methods is created and a coordinate descent method for linear SVM to Maxent is extended, showing that it is faster than existing iterative scaling methods.
- MetaDistiller: Network Self-Boosting via Meta-Learned Top-Down Distillation44
This work proposes that better soft targets with higher compatibil-ity can be generated by using a label generator to fuse the feature maps from deeper stages in a top-down manner, and can employ the meta-learning technique to optimize this label generator.
- FINGER: Fast Inference for Graph-based Approximate Nearest Neighbor Search43
FINGER is proposed, a fast inference method for efficient graph search in AKNNS that approximates the distance function by estimating angles between neighboring residual vectors and can be used to bypass unnecessary computations for faster searches.
- 43
- 40
- Asynchronous Parallel Greedy Coordinate Descent40
The proposed Asynchronous parallel Greedy Coordinate Descent algorithm for minimizing a smooth function with bounded constraints enjoys good speedup and outperforms existing multi-core kernel SVM solvers including asynchronous stochastic coordinate descent and multi- core LIBSVM.
- Improving the Adversarial Robustness of NLP Models by Information Bottleneck39
Through extensive experiments, it is shown that the models trained with the information bottleneck-based method are able to achieve a significant improvement in robust accuracy, exceeding performances of all the previously reported defense methods while suffering almost no performance drop in clean accuracy on SST-2, AGNEWS and IMDB datasets.
- 39
- Large Scale Distributed Sparse Precision Estimation39
This work considers the problem of sparse precision matrix estimation in high dimensions using the CLIME estimator, and presents an inexact alternating direction method of multiplier (ADMM) algorithm for CLIME, and develops a large scale distributed framework for the computations.
- Random Sharpness-Aware Minimization38
R-SAM essentially smooths the loss landscape, based on which it is able to apply the one-step gradient ascent on the smoothed weights to improve the approximation of the inner maximization.
- 38
- 38
- RedSync: Reducing synchronization bandwidth for distributed deep learning training system37
This paper develops an RGC method that is able to reduce the end-to-end training time on real-world multi-GPU systems, and introduces a set of optimizations to reduce communication bandwidth while introducing limited overhead.
- 37
- On Discrete Prompt Optimization for Diffusion Models36
Empirical evaluation on prompts collected from diverse sources suggests that the first gradient-based framework for prompt optimization in text-to-image diffusion models can discover prompts that substantially improve or destroy the faithfulness of images generated by the text-to-image diffusion model.
- LoRA Done RITE: Robust Invariant Transformation Equilibration for LoRA Optimization36
LoRA-RITE is introduced, a novel adaptive matrix preconditioning method for LoRA optimization, which can achieve transformation invariance and remain computationally efficient.
- On the Sensitivity and Stability of Model Interpretations in NLP36
A new class of interpretation methods that adopt techniques from adversarial robustness, motivated by the desiderata of sensitivity and stability, are introduced that are effective under the new criteria and overcome limitations of gradient-based methods on removal-based criteria.
- On the Transferability of Adversarial Attacks against Neural Text Classifier35
This paper presents the first study to systematically investigate the transferability of adversarial examples for text classification models and proposes a genetic algorithm to find an ensemble of models that can be used to induce adversarialExamples to fool almost all existing models.
- Improved Adversarial Training via Learned Optimizer35
It is empirically demonstrate that the commonly used PGD attack may not be optimal for inner maximization, and improved inner optimizer can lead to a more robust model.
- 35
- MOSSBench: Is Your Multimodal Language Model Oversensitive to Safe Queries?34
It is demonstrated that advanced Multimodal Large Language Models (MLLMs) exhibit similar tendencies to safety mechanisms, highlighting the need for refined safety mechanisms that balance caution with contextually appropriate responses, improving the reliability of MLLMs in real-world applications.
- Investigating heterogeneities of live mesenchymal stromal cells using AI-based label-free imaging34
An artificial intelligence (AI)-based method that converts transmitted light microscopy images of MSCs into quantitative measurements of protein expression levels and shows that expression of surface markers provides a heterogeneity characterization that is complementary to conventional cell-level morphological analyses is developed.
- LoL: Longer than Longer, Scaling Video Generation to Hour33
This work proposes a lightweight, training-free approach that effectively suppresses sink-collapse by introducing multi-head RoPE jitter that breaks inter-head attention homogenization and mitigates long-horizon collapse, and achieves the first demonstration of real-time, streaming, and infinite-length video generation with little quality decay.
- Extreme Zero-Shot Learning for Extreme Text Classification33
A pre-training method, MACLR, which thoroughly leverages the raw text with techniques including \textbf{M}ulti-scale, which achieves superior performance compared to all other leading baseline methods, in particular with approximately 5-10% improvement in precision and recall on average.
- Generalizing Few-Shot NAS with Gradient Matching33
A gradient matching score (GM) that leverages gradient information at the shared weight for making informed splitting decisions is proposed that significantly outperforms its Few-Shot counterparts while surpassing previous comparable methods in terms of the accuracy of derived architectures.
- Training Meta-Surrogate Model for Transferable Adversarial Attack32
Comprehensive experiments on Cifar-10 and ImageNet demonstrate that by attacking the MSM, one can obtain stronger transferable adversarial examples to deceive black-box models including adversarially trained ones, with much higher success rates than existing methods.
- Robust principal component analysis with side information32
This paper proposes a convex problem to incorporate side information in robust PCA and shows that the low rank matrix can be exactly recovered via the proposed method under certain conditions.
- 30
- 30
- 30
- One Prompt is not Enough: Automated Construction of a Mixture-of-Expert Prompts29
The Mixture-of-Expert paradigm is adopted and the resulting method, codenamed Mixture-of-Prompts (MoP), achieves an average win rate of 81% against prior arts across several major benchmarks.
- 29
- 29
- 29
- 28
- Two-stage LLM Fine-tuning with Less Specialization and More Generalization28
This work proposes Prompt Tuning with MOdel Tuning (ProMoT), a simple yet effective two-stage fine-tuning framework that preserves in-context abilities of the pretrained model and shows remarkable generalization ability on tasks that have different formats.
- 28
- Sparse Linear Programming via primal and dual augmented coordinate descent28
This paper investigates a general LP algorithm based on the combination of Augmented Lagrangian and Coordinate Descent (AL-CD), giving an iteration complexity of O((log(1/∊))2) with O(nnz(A)) cost per iteration, and yields a tractable alternative to standard LP methods for large-scale problems of sparse solutions and nnz(A) ≪ mn.
- 27
- 27
- 27
- Concept Gradient: Concept-based Interpretation Without Linear Assumption26
This work showed that for a general (potentially non-linear) concept, it can be mathematically evaluated how a small change of concept affecting the model's prediction leads to an extension of gradient-based interpretation to the concept space.
- 26
- 25
- Learning from Group Comparisons: Exploiting Higher Order Interactions25
This paper proposes a latent factor model, and shows that the sample complexity of the model is bounded under mild assumptions, and has much better prediction power on several E-sports datasets, and furthermore can be used to reveal interesting patterns that cannot be discovered by previous methods.
- A Fast Sampling Algorithm for Maximum Inner Product Search24
Sampling-MIPS is proposed, which is the first sampling based algorithm that can be applied to the MIPS problem on a set of general vectors with both positive and negative values, and is efficient in terms of both time and sample complexity.
- 23
- 23
- Constant Nullspace Strong Convexity and Fast Convergence of Proximal Methods under High-Dimensional Settings23
This paper investigates a novel variant of strong convexity, which it is called Constant Nullspace Strong Convexity (CNSC), where the objective function be strongly convex only over a constant subspace.
- Clustering and Constructing User Coresets to Accelerate Large-scale Top-K Recommender Systems22
A novel model for clustering and navigating for top-K recommenders (CANTOR) is proposed to expedite the computation of top- K recommendations based on latent factor models and can significantly reduce the computation without compromising the quality of the recommendations.
- 22
- Learning Word Embeddings for Low-Resource Languages by PU Learning22
A Positive-Unlabeled Learning (PU-Learning) approach is designed to factorize the co-occurrence matrix and validate the proposed approaches in four different languages.
- The Crystal Ball Hypothesis in diffusion models: Anticipating object positions from initial noise21
This study identifies specific regions within the initial noise image that play a key role for object generation in the resulting images, termed trigger patches, and proposes a reject-sampling strategy to obtain optimal noise, aiming to improve prompt adherence and positional diversity in image generation.
- Phenotyping senescent mesenchymal stromal cells using AI image translation20
It is demonstrated that AI image translation algorithms can effectively predict immunofluorescence images of MSC senescence markers from phase contrast images and the expression level of senescenced markers including senescENCE-associated beta-galactosidase (SABG), p16, p21, and p38 are accurately predicted by deep-learning models.
- Large-scale Collaborative Ranking in Near-Linear Time20
This is the first collaborative ranking algorithm capable of working on the full Netflix dataset using all the 20 billion rating pairs, and this leads to a model with much better recommendation compared with previous models trained on subsamples.
- Organizational overlap on social networks and its applications20
This paper addresses the problem of computing edge affinity between two users on a social network, based on the users belonging to organizations such as companies, schools, and online groups, and presents a novel mathematical model to compute the probability of connection between two people based on organizational overlap.
- Local Critic Training for Model-Parallel Learning of Deep Neural Networks19
A novel model-parallel learning method which trains neural networks using additional modules called local critic networks, which successfully decouples the update process of the layer groups for both convolutional neural Networks (CNNs) and recurrent neural networks (RNNs).
- 19
- 19
- 18
- 18
- Data Attribution for Diffusion Models: Timestep-induced Bias in Influence Estimation17
This work introduces Diffusion-ReTrac as a re-normalized adaptation that enables the retrieval of training samples more targeted to the test sample of interest, facilitating a localized measurement of influence and considerably more intuitive visualization.
- 17
- An ensemble of three classifiers for KDD cup 2009: expanded linear model, heterogeneous boosting, and selective naïve Bayes17
This paper transforms the three binary classification tasks into a joint multi-class classification problem, and solves an l1-regularized maximum entropy model under the LIBLINEAR framework, and proposes a heterogeneous base learner, which is capable of handling different types of features and missing values.
- RANK-NOSH: Efficient Predictor-Based Architecture Search via Non-Uniform Successive Halving16
This work proposes NOn-uniform Successive Halving (NOSH), a hierarchical scheduling algorithm that terminates the training of underperforming architectures early to avoid wasting budget, and forms predictor-based architecture search as learning to rank with pairwise comparisons.
- Automatic Perturbation Analysis on General Computational Graphs.16
The main idea is to express a network as a computational graph and then generalize linear relaxation algorithms such as CROWN as a graph algorithm, and its computation can be done automatically in a similar manner as the back-propagation algorithm for gradient computation.
- Improved Bounded Matrix Completion for Large-Scale Recommender Systems16
This paper proposes the new approach for solving BMC under the ADMM framework, which can reach a lower objective function value, obtain a higher prediction accuracy and have better scalability compared with existing bounded matrix completion approaches.
- 15
- PECOS15
This tutorial will introduce several key functions and features of the PECOS library, and learn how to efficiently train large-scale machine learning models for enormous output spaces, and obtain predictions in less than 1 millisecond for a data input with million labels.
- 15
- 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.
- Iterative scaling and coordinate descent methods for maximum entropy15
This paper creates a general and unified framework for IS methods, this framework also connects IS and coordinate descent (CD) methods, and develops a CD method for Maxent.
- 14
- Double Perturbation: On the Robustness of Robustness and Counterfactual Bias Evaluation14
A “double perturbation” framework is proposed to uncover model weaknesses beyond the test dataset and is able to reveal the hidden model biases not directly shown in the test datasets.
- 14
- 14
- 14
- JIGMARK: A Black-Box Approach for Enhancing Image Watermarks against Diffusion Model Edits13
JIGMARK is introduced, a first-of-its-kind watermarking technique that enhances robustness through contrastive learning with pairs of images, processed and unprocessed by diffusion models, without needing a direct backpropagation of the diffusion process.
- 13
- NeSSA: Near-Storage Data Selection for Accelerated Machine Learning Training13
NeSSA is proposed, a novel SmartSSD+GPU training architecture to intelligently select important subsets of large datasets near-storage, such that training on a subset mimics training on the full dataset with a very small loss in accuracy.
- 13
- Fast Training for Large-Scale One-versus-All Linear Classifiers using Tree-Structured Initialization13
A novel algorithm called OVA-Primal++ is proposed that speeds up the training of OVA by using a tree-structured training order, where each classifier is trained using its parent’s classifier as initialization.
- Temporal shuffling for defending deep action recognition models against adversarial attacks12
It is suggested that action recognition models rely on the motion information less than expected, and thus they are robust to randomization of frame orders, and it is found that motion monotonicity remaining after randomization also contributes to such robustness.
- Can Agents Run Relay Race with Strangers? Generalization of RL to Out-of-Distribution Trajectories12
A novel method called Self-Trajectory Augmentation (STA), which will reset the environment to the agent's old states according to the Q function during training, is proposed, which reduces the failure rate of SAC under relay-evaluation by more than three times in most settings without impacting agent performance and increasing the needed number of environment interactions.
- PINA: Leveraging Side Information in eXtreme Multi-label Classification via Predicted Instance Neighborhood Aggregation12
This work proposes Predicted Instance Neighborhood Aggregation (PINA), a data enhancement method for the general XMC problem that leverages beneficial side information from the label metadata and the correlations among training instances.
- 12
- 12
- Spanning attack: reinforce black-box attacks with unlabeled data12
By constraining adversarial perturbations in a low-dimensional subspace via spanning an auxiliary unlabeled dataset, the spanning attack significantly improves the query efficiency of a wide variety of existing black-box attacks.
- 12
- 12
- Computationally efficient Nyström approximation using fast transforms12
This paper proposes the use of a family of fast transforms to generate structured landmark points for Nystrom approximation by exploiting fast transforms, e.g., Haar transform and Hadamard transform, which can significantly speed up kernel approximation and benefit prediction speed in kernel machines.
- A divide-and-conquer procedure for sparse inverse covariance estimation12
A clustering algorithm is proposed that attempts to minimize the quality of the approximate solution obtained from solving the corresponding sub-divided problems, and uses the solutions of the sub-problems to build a good approximation for the original problem.
- Large Language Models are Interpretable Learners11
Empirical results demonstrate LSP's superior performance compared to traditional neurosymbolic programs and vanilla automatic prompt tuning methods, and as the knowledge learned by LSP is a combination of natural language descriptions and symbolic rules, it is easily transferable to humans, and other LLMs, and generalizes well to out-of-distribution samples.
- Gandalf: Learning Label-label Correlations in Extreme Multi-label Classification via Label Features11
Gandalf is proposed, a novel approach which makes use of a label co-occurrence graph to leverage label features as additional data points to supplement the training distribution and exploit the characteristics of the short-text XMC problem.
- Improving the Generation Quality of Watermarked Large Language Models via Word Importance Scoring11
This work proposes to improve the quality of texts generated by a watermarked language model by Watermarking with Importance Scoring (WIS), and proposes three methods to predict importance scoring, including a perturbation-based method and two model-based methods.
- On the Convergence of Certified Robust Training with Interval Bound Propagation11
It is shown that when using IBP training to train a randomly initialized two-layer ReLU neural network with logistic loss, gradient descent can linearly converge to zero robust training error with a high probability if the authors have sufficiently small perturbation radius and large network width.
- 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.
- Multi-Proxy Wasserstein Classifier for Image Classification11
This work adopts optimal transport theory to calculate a non-uniform matching flow between the elements in the feature map of a sample and the proxies of a class in a closed way, and enables the models to achieve partial matching.
- 11
- 11
- Multiple Accounts Detection on Facebook Using Semi-Supervised Learning on Graphs11
This paper proposes using graph embedding together with semi-supervised learning to predict whether multiple accounts are created by the same user, and notes that local information have higher importance than the global ones for such prediction, and points out the threshold leading to the best results.
- MuLan: Multimodal-LLM Agent for Progressive and Interactive Multi-Object Diffusion10
A training-free Mu ltimodal-L LM (MuLan) to address challenges by progressive multi-object generation with planning and feedback control, like a human painter, demonstrating the superiority of MuLan in generating multiple objects over baselines.
- Towards Adversarially Robust Text Classifiers by Learning to Reweight Clean Examples10
This study proposes to train text classifiers by a sample reweighting method in which the example weights are learned to minimize the loss of a validation set mixed with the clean examples and their adversarial ones in an online learning manner.
- 2.5D visual relationship detection10
This work studies 2.5D visual relationship detection (2.5VRD), in which the goal is to jointly detect objects and predict their relative depth and occlusion relationships, and shows that existing models largely rely on semantic cues and simple heuristics to solve this task.
- Deep Image Destruction: Vulnerability of Deep Image-to-Image Models against Adversarial Attacks10
It is shown that unlike image classification tasks, the performance degradation on image-to-image tasks largely differs depending on various factors, e.g., attack methods and task objectives.
- QUIC & DIRTY: A Quadratic Approximation Approach for Dirty Statistical Models10
A family of algorithms for optimizing "superposition-structured" or "dirty" statistical estimators for high-dimensional problems involving the minimization of the sum of a smooth loss function with a hybrid reg-ularization is developed.
- Adversarial Examples Detection With Bayesian Neural Network9
A novel Bayesian adversarial example detector is proposed, short for BATer, to improve the performance of adversarial example detection and use the randomness of the Bayesian neural network to simulate hidden layer output distribution and leverage the distribution dispersion to detect adversarial examples.
- Trustworthy in silico cell labeling via ensemble-based image translation9
It is demonstrated that an ensemble learning method can quantify the uncertainty of AI image translation, and suggested that the ensemble-estimated uncertainty can be a useful indicator for identifying erroneous AI image translations.
- 9
- 9
- Adversarially Robust Deep Image Super-Resolution Using Entropy Regularization9
This paper proposes a defense method that is formulated as an entropy regularization loss for model training, which can be augmented to the original training loss of super-resolution models, the first attempt of adversarial defense for deep super-resolution models.
- Learning to Stop: Dynamic Simulation Monte-Carlo Tree Search9
The algorithm, called Dynamic Simulation MCTS (DS-MCTS), is proposed, which can speed up a NoGo agent trained by AlphaZero 2.5 times faster while maintaining a similar winning rate, which is critical for training and conducting experiments.
- 9
- 9
- ClawEnvKit: Automatic Environment Generation for Claw-Like Agents8
ClawEnvKit is introduced, an autonomous generation pipeline that instantiates this formalism from natural language descriptions and enables live evaluation: users describe a desired capability in natural language and obtain a verified environment on demand, turning evaluation into a continuous, user-driven process.
- Low-rank Matrix Bandits with Heavy-tailed Rewards8
A novel algorithm called LOTUS attaining the regret bound of order $\tilde O(d^\frac{3}{2}r^\frac{1}{1+\delta}/\tilde{D}_{rr})$ without knowing $T$ is proposed, which matches the state-of-the-art regret bound under sub-Gaussian noises.
- Supracellular measurement of spatially varying mechanical heterogeneities in live monolayers8
A monolayer effective modulus measurement is developed by integrating a custom cell stretcher, light microscopy, and AI-based inference and observed a mild correlation between cell area and stiffness in jammed epithelia, suggesting the influence of cell modulus on packing.
- 8
- 8
- 8
- 8
- 8
- 8
- 8
- An Efficient Rehearsal Scheme for Catastrophic Forgetting Mitigation during Multi-stage Fine-tuning7
A sampling scheme that prioritizes rehearsal of ``collateral damage'' samples, which are samples predicted correctly by the prior model but forgotten by the incrementally tuned one, and which outperforms several leading continual learning methods in compute-constrained settings.
- 7
- Mitigating Bias in Dataset Distillation7
To reduce bias amplification in dataset condensation, a simple yet highly effective approach based on a sample reweighting scheme utilizing kernel density estimation is introduced, demonstrating the effectiveness of the proposed method on multiple real-world and synthetic datasets.
- 7
- Fast LSTM by dynamic decomposition on cloud and distributed systems7
This paper designs and implements a distributed optimization system with dynamic decomposition, which can significantly reduce the energy cost and accelerate the training process and achieves average speedup for six real-world applications without losing accuracy in inference.
- 7
- 7
- From Adversarial Training to Generative Adversarial Networks7
A system that combines generator, discriminator, and adversarial attacker in a single network that can simultaneously improve the robustness of classifiers, measured by accuracy under strong adversarial attacks; and the quality of generators, evaluated both aesthetically and quantitatively is proposed.
- 7
- Communication-Efficient Distributed Block Minimization for Nonlinear Kernel Machines7
This paper develops a parallel block minimization framework, and demonstrates its good scalability in solving nonlinear kernel SVM and logistic regression, and proves global linear convergence rate of the proposed method with a wide class of subproblem solvers.
- Entity Disambiguation with Extreme Multi-label Ranking6
A novel framework, eXtreme Multi-label Ranking for Entity Disambiguation (XMRED), which outperforms the previous state-of-the-art for page-level document retrieval and applies deep cross-encoder as a re-ranker to achieve high precision based on high-quality candidates.
- Block Low-Rank Preconditioner with Shared Basis for Stochastic Optimization6
This work proposes approximating the diagonal blocks of the second moment matrix by low-rank matrices and enforcing the same basis for the blocks within each layer within each layer, and designs an algorithm to maintain this shared-basis block low-rank approximation during training.
- Stochastic Optimization for Nonconvex Problem With Inexact Hessian Matrix, Gradient, and Function6
This article explores a family of stochastic TR (STR) and stochastic ARC (SARC) methods that can simultaneously provide inexact computations of the Hessian matrix, gradient, and function values and proves that the iteration complexity to achieve approximating second-order optimality is of the same order as the exact computations demonstrated in previous studies.
- 6
- Relevance under the Iceberg6
A novel framework, Model-Agnostic List Truncation with Ordinal Regression (MALTOR), is proposed to leverage the ranking properties and truncate long ranked label lists for better accuracy and to provide reasonable prediction for extreme multi-label classification with dynamic numbers of predicted labels.
- 6
- Efficient Contextual Representation Learning With Continuous Outputs6
This work revisits the design of the output layer and considers directly predicting the pre-trained embedding of the target word for a given context and achieves a 4-fold speedup and eliminates 80% trainable parameters while achieving competitive performance on downstream tasks.
- Rank Aggregation and Prediction with Item Features6
This work proposes a general model to learn a total ranking by balancing between comparisons and feature information jointly and shows that given sufficiently informative features, the sample complexity of the model can be asymptotically lower than models based only on comparisons for deriving an accurate ranking.
- Representer Point Selection for Explaining Regularized High-dimensional Models5
The workhorse is a novel representer theorem for general regularized high-dimensional models, which decomposes the model prediction in terms of contributions from each of the training samples: with positive (negative) values corresponding to positive ( negative) impact training samples to the model's prediction.
- Uncertainty Quantification for Extreme Classification5
This work analyzes label-level and instance-level uncertainty in eXtreme Multi-label Classification (XMC), and proposes a general approximation framework based on beam search to efficiently estimate the uncertainty with a theoretical guarantee under long-tail XMC predictions.
- 5
- 5
- 5
- Communication-avoiding kernel ridge regression on parallel and distributed systems5
The balanced K RR (BKRR) and K-means KRR (KKRR) are proposed, which consider alternative ways to partition the input dataset into pdifferent parts, generating pdifferent models, and then selecting the best model among them.
- 5
- 5
- UniDEC : Unified Dual Encoder and Classifier Training for Extreme Multi-Label Classification4
UniDEC is developed, a loss-independent, end-to-end trainable framework which trains the DE and classifier together in a unified manner with a multi-class loss, while reducing the computational cost by 4--16x.
- POME: Post Optimization Model Edit via Muon-style Projection4
This work introduces Post-Optimization Model Edit (POME), a new algorithm that enhances the performance of fine-tuned large language models using only their pretrained and fine-tuned checkpoints, without requiring extra data or further optimization.
- CLUE: Concept-Level Uncertainty Estimation for Large Language Models4
A novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs is proposed, which can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.
- PEFA: Parameter-Free Adapters for Large-scale Embedding-based Retrieval Models4
This work proposes the PEFA framework, namely ParamEter-Free Adapters, for fast tuning of ERMs without any backward pass in the optimization, and achieves significant improvement on two retrieval applications.
- Certified Training with Branch-and-Bound for Lyapunov-stable Neural Control4
A novel framework of training-time branch-and-bound to dynamically maintain a training dataset of subregions throughout training, such that the hardest subregions are iteratively split into smaller ones whose verified bounds can be computed more tightly to ease the training.
- 4
- 4
- 4
- 4
- 4
- 4
- Extreme Learning to Rank via Low Rank Assumption4
A fast algorithm is developed to reduce the time complexity of gradient descent solver by exploiting the low-rank structure and the resulting algorithm is much faster than existing methods.
- A comparison of second-order methods for deep convolutional neural networks4
Although currently second-order methods are too slow to be applied in practice, it can reduce training loss in fewer number of iterations compared with SGD, and quasi-newton methods with more self-adjusting mechanisms might be more useful than SGD in training deeper networks.
- 4
- 3
- 3
- 3
- 3
- 3
- Provable, Scalable and Automatic Perturbation Analysis on General Computational Graphs3
This work develops an automatic framework to enable perturbation analysis on any neural network structures, by generalizing exiting LiRPA algorithms such as CROWN to operate on general computational graphs and yields an open-source library for the community to applyLiRPA to areas beyond certified defense without much LiR PA expertise.
- 3
- 3
- 3
- ImageNet Training by CPU: AlexNet in 11 Minutes and ResNet-50 in 48 Minutes3
It is confirmed that LARS could be used to further scale the number of processors efficiently used in DNN training and, and as a result, further reduce the total training time.
- 3
- DualEval: Joint Model-Item Calibration for Unified LLM Evaluation2
This work introduces DualEval, a latent model-item calibration framework that represents models and evaluation items in a shared space, jointly estimating model ability together with item difficulty and sharpness, and unifies static and arena-style evaluation.
- Matryoshka Model Learning for Improved Elastic Student Models2
This paper proposes MatTA, a framework for training multiple accurate Student models using a novel Teacher-TA-Student recipe, which provides multiple servable options to trade off accuracy for lower serving cost.
- 2
- 2
- Expert Proximity as Surrogate Rewards for Single Demonstration Imitation Learning2
TDIL is an IRL method designed to address reward sparsity by introducing a denser surrogate reward function that considers environmental dynamics and trains a transition discriminator to differentiate between valid and non-valid transitions in a given environment to compute the surrogate rewards.
- Build Faster with Less: A Journey to Accelerate Sparse Model Building for Semantic Matching in Product Search2
This paper investigates and accelerate the model building procedures in a tree-based XMC model and achieves over 10 times acceleration on a real-world semantic matching task with 100M labels.
- Online Continuous Hyperparameter Optimization for Generalized Linear Contextual Bandits2
This work uses a double-layer bandit framework named CDT (Continuous Dynamic Tuning) and forms the hyperparameter optimization as a non-stationary continuum-armed bandit, where each arm represents a combination of hyperparameters, and the corresponding reward is the algorithmic result.
- 2
- 2
- 2
- Uncertainty in Extreme Multi-label Classification2
This work analyzes label-level and instance-level uncertainty in eXtreme Multi-label Classification (XMC), and proposes a general approximation framework based on beam search to efficiently estimate the uncertainty with a theoretical guarantee under long-tail XMC predictions.
- 2
- 2
- Generating universal language adversarial examples by understanding and enhancing the transferability across neural models2
This paper systematically study the transferability of adversarial attacks for text classification models and proposes universal black-box attack algorithms that can induce adversarial examples to attack almost all existing models.
- 2
- 2
- 2
- 2
- 2
- Parallel Asynchronous Stochastic Coordinate Descent with Auxiliary Variables2
A rich family of objective functions where AUX-PCD can be applied is considered and theoretical guarantees for many practical parallel coordinate descent implementations, which currently lack guarantees are provided.
- 2
- Uncertainty‐Guided Selective Adaptation Enables Cross‐Platform Predictive Fluorescence Microscopy1
SIT‐ADDA with automatic depth selection (SIT‐ADDA‐Auto), a self‐configuring framework that integrates shallow‐layer adversarial alignment with predictive uncertainty to automatically select adaptation depth without target labels, demonstrates robustness via multimetric evaluation, blinded expert assessment, and uncertainty‐depth ablations.
- Arena-T2I Hard: Benchmarking and Improving Faithfulness with Dependency-Aware Checklist1
A dependency-aware checklist reward is proposed that decomposes each prompt into a DAG of yes/no questions and zeroes descendants of failed parents, turning faithfulness into a per-constraint training signal, which attains a strictly better faithfulness-aesthetics trade-off on SD3.5-Medium and FLUX.
- QG-CoC: Question-Guided Chain-of-Captions for Large Multimodal Models1
This work extensively investigates how current prompting methods perceive fine-grained visual details and process visual information when dealing with multiple images and proposes a new zero-shot prompting method, Question-Guided Chain-of-Captions (QG-CoC), a generalized prompting approach that effectively handles problems with an arbitrary number of images.
- 1
- A refined reweighing technique for nondiscriminatory classification1
A novel data pre-processing technique that assigns weights to training instances in order to reduce discrimination without changing any of the inputs or labels is proposed.
- Fast block-wise partitioning for extreme multi-label classification1
Experiments on benchmark multi-label data sets reveal that BP pretreatment significantly reduces prediction time, and retains almost the same level of prediction accuracy.
- 1
- 1
- 1
- 1
- Learning to Learn with Smooth Regularization1
Comprehensive experiments on the neural network training tasks demonstrate that the proposed regularization consistently improve the learned neural optimizers even when transferring to tasks with different architectures and datasets.
- 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.
- Measures and Best Practices for Responsible AI1
This workshop invites contributions towards different datasets which help enhance task performance and inclusivity, measures and metrics which help in determining the trustworthiness of a model/dataset, assessment or remediation tools for fairer, more transparent, robust, and reliable models, and case studies describing responsible development of AI systems across fields such as healthcare, financial services, insurance, etc.
- 1
- 1
- 1
- MulCode: A Multiplicative Multi-way Model for Compressing Neural Language Model1
This work proposes MulCode, a novel multi-way multiplicative neural compressor that learns an adaptively created matrix and its multiplicative compositions and is more effective than the state-of-the-art compression methods.
- 1
- Positive-Unlabeled Demand-Aware Recommendation.1
This work proposes a highly scalable alternating minimization approach to recommendation for e-commerce with a mix of durable and nondurable goods that can solve problems with millions of users and items and shows superior prediction accuracies on multiple real-world data sets.
- Mechanism‐Informed Language Modeling and Oxygenated 3D Screening Identify Berberine–Enzalutamide Synergy in Prostate Cancer Models–
This work introduces a hybrid in silico and in vitro lead discovery platform that integrates knowledge‐augmented large language models (LLMs) with an oxygen‐supplemented 3D spheroid system and establishes proof of concept that mechanism‐informed language modeling coupled with 3D screening can prioritize and experimentally evaluate drug‐combination hypotheses at an early stage of discovery.
- –
- Closing the Auto-Research Loop: An AI Co-Scientist for Production Search Ranking–
It is demonstrated that this framework discovered a novel technique for handling sequence features, with all model enhancements produced automatically, yielding substantial offline performance improvements and suggesting that AI systems can discover ranking architectures comparable to those developed by human experts while significantly reducing routine research workloads.
- Differentiable Efficient Operator Search–
This work introduces Efficient Operator Search, a differentiable framework that jointly searches where to reduce tokens, how many tokens to retain, and how reduced token information should be processed, and suggests that efficient multimodal inference can be reframed from manual operator design to differentiable operator search.
- –
- FRESCO: Benchmarking and Optimizing Re-rankers for Evolving Semantic Conflict in Retrieval-Augmented Generation–
This work introduces FRESCO (Factual Recency and Evolving Semantic COnflict), a benchmark for evaluating re-rankers in temporally dynamic contexts, and identifies Pareto-optimal instructions that balance Evolving and Non-Evolving Knowledge tasks.
- MIRAGE: Protecting against Malicious Image Editing via False Moderation–
This work proposes to immunize images by causing pre-generation safety moderation classifiers to flag images as policy-violating, triggering an automatic refusal regardless of the editing prompt, and calls this method MIRAGE, which stands for Moderation Induced Resistance Against Generative Editing.
- –
- –
- –
- Text is All You Need for Vision-Language Model Jailbreaking–
This work introduces Text-DJ, a novel jailbreak attack that bypasses LVLMs' OCR capabilities that are not robust to dispersed, multi-image adversarial inputs, highlighting the need for defenses for fragmented multimodal inputs.
- –
- –
- Compressing Many-Shots in In-Context Learning–
This work systematically evaluates various compressor models and training approaches across different model sizes, architectures, many-shot sequence lengths, and compression ratios, and proposes MemCom, a layer-wise compression method that outperforms strong baselines across all compression ratios on multiple classification tasks with large label sets.
- –
- –
- Provably Robust Training of Quantum Circuit Classifiers Against Parameter Noise–
This work presents a provably noise-resilient training theory and algorithm to enhance the robustness of parameterized quantum circuit classifiers, and guarantees resilience to parameter noise with minimal adjustments to commonly used optimization algorithms.
- Randomized Benchmarking of Local Zeroth-Order Optimizers for Variational Quantum Systems–
This work compares the performance of classical optimizers across a series of partially-randomized tasks to more broadly sample the space of variational quantum algorithm problems and focuses on local zeroth-order optimizers due to their generally favorable performance and query-efficiency on quantum systems.
- –
- UNICORN: A Unified Causal Video-Oriented Language-Modeling Framework for Temporal Video-Language Tasks–
This paper adopts visual instruction tuning to build a unified causal video-oriented language modeling framework, named UNICORN, and shows that training on the generated captions improves the performance of video-language models on both zero-shot and fine-tuning settings.
- –
- Embedding Space Selection for Detecting Memorization and Fingerprinting in Generative Models–
This method capitalizes on the unique distributions of the memorization score across different layers of ViTs, providing a novel approach to identifying models involved in generating deepfakes and malicious content and demonstrates a marked 30% enhancement in identification accuracy over existing baseline methods, offering a more effective tool for combating digital misinformation.
- Enhancing Unsupervised Semantic Parsing with Distributed Contextual Representations–
A non-parametric Bayesian model is extended to deal with homonymy and polysemy by leveraging distributed contextual word and phrase representations pre-trained on a large collection of unla-belled texts and achieves better accuracy on both question-answering and relation extraction tasks.
- Federated Learning with Projected Trajectory Regularization–
This paper proposes a novel federated learning framework with projected trajectory regularization (FedPTR) for tackling the data heterogeneity issue, which proposes a unique way to better extract the essential global information from the model training trajectory.
- –
- –
- –
- –
- –
- –
- –
- 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.
- –
- –
- –
- –
- Improving Adversarial Robustness to Sensitivity and Invariance Attacks with Deep Metric Learning–
The preliminary results indicate that regularizing over invariant perturbations in the framework improves both invariant and sensitivity defense and promotes robustness toward both types of attacks beyond Euclidean distance metrics.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Toward Finding The Global Optimal of Adversarial Examples–
A general boosting algorithm that can help current attack to find a more global optimal example and reduce the distortion between the original image and adversarial sample about 10%-20% by adopting dynamic successive halving.
- –
- –
- –
- 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.
- –
- Distributed Primal-Dual Optimization for Non-uniformly Distributed Data–
This work develops a computational efficient algorithm to automatically choose the optimal weights for each machine in the primal-dual optimization framework and proposes an efficient way to estimate the duality gap of the merged update by exploiting the structure of the objective function.
- –
- Better Generalization by Efficient Trust Region Method–
A trust region method for training deep neural networks that only requires Hessian-vector products, which can be computed efficiently by back-propagation in neural networks, and finds a model that has lower generalization (test) error than SGD, and this difference is even more significant in large batch training.
- –
- Machine Learning Meliorates Computing and Robustness in Discrete Combinatorial Optimization Problems–
This work demonstrates computational advantages and intrinsic robustness in the two systems, and proposes brand new robust optimal solutions that are potentially more realistic and practical in real world settings.
- –
- –
- –
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 2026-10-15. 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.