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
- Jason J. CorsoSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works458 from public data
- Knowing When to Look: Adaptive Attention via a Visual Sentinel for Image Captioning1,670
This paper proposes a novel adaptive attention model with a visual sentinel that sets the new state-of-the-art by a significant margin on image captioning.
- Large language models generate functional protein sequences across diverse families1,189
ProGen is described, a language model that can generate protein sequences with a predictable function across large protein families, akin to generating grammatically and semantically correct natural language sentences on diverse topics.
- 1,137
- SummEval: Re-evaluating Summarization Evaluation1,112
This work re-evaluate 14 automatic evaluation metrics in a comprehensive and consistent fashion using neural summarization model outputs along with expert and crowd-sourced human annotations and implements and shares a toolkit that provides an extensible and unified API for evaluating summarization models across a broad range of automatic metrics.
- Evaluating the Factual Consistency of Abstractive Text Summarization1,066
A weakly-supervised, model-based approach for verifying factual consistency and identifying conflicts between source documents and a generated summary substantially outperforms previous models, including those trained with strong supervision using standard datasets for natural language inference and fact checking.
- ERASER: A Benchmark to Evaluate Rationalized NLP Models957
This work proposes the Evaluating Rationales And Simple English Reasoning (ERASER) a benchmark to advance research on interpretable models in NLP, and proposes several metrics that aim to capture how well the rationales provided by models align with human rationales, and also how faithful these rationales are.
- Diffusion Model Alignment Using Direct Preference Optimization923
This work proposes Diffusion-DPO, a method to align diffusion models to human preferences by directly optimizing on human comparison data, and fine-tuned the base model of the state-of-the-art Stable Diffusion XL (SDXL)-1.0 model with Diffusion-DPO.
- 837
- Unified Training of Universal Time Series Forecasting Transformers831
This work presents novel enhancements to the conventional time series Transformer architecture, resulting in the proposed Masked Encoder-based Universal Time Series Forecasting Transformer (Moirai), which achieves competitive or superior performance when compared to full-shot models.
- 731
- 730
- Explain Yourself! Leveraging Language Models for Commonsense Reasoning663
This work collects human explanations for commonsense reasoning in the form of natural language sequences and highlighted annotations in a new dataset called Common Sense Explanations to train language models to automatically generate explanations that can be used during training and inference in a novel Commonsense Auto-Generated Explanation framework.
- End-to-End Dense Video Captioning with Masked Transformer612
This work proposes an end-to-end transformer model, which employs a self-attention mechanism, which enables the use of efficient non-recurrent structure during encoding and leads to performance improvements.
- 603
- A Joint Many-Task Model: Growing a Neural Network for Multiple NLP Tasks589
A joint many-task model together with a strategy for successively growing its depth to solve increasingly complex tasks and uses a simple regularization term to allow for optimizing all model weights to improve one task’s loss without exhibiting catastrophic interference of the other tasks.
- A comprehensive survey on pretrained foundation models: a history from BERT to ChatGPT556
This survey aims to shed light on the research of the PFMs on scalability, security, logical reasoning ability, cross-domain learning ability, cross-domain learning ability, and user-friendly interactive ability for artificial general intelligence.
- Intent Contrastive Learning for Sequential Recommendation509
Intent Contrastive Learning (ICL) is proposed, a general learning paradigm that leverages a latent variable to represent users’ intents and learn the distribution function of the latent variable via clustering in SR models via contrastive SSL.
- Transferable Multi-Domain State Generator for Task-Oriented Dialogue Systems497
A Transferable Dialogue State Generator (TRADE) that generates dialogue states from utterances using copy mechanism, facilitating transfer when predicting (domain, slot, value) triplets not encountered during training.
- 456
- 454
- 450
- ULIP: Learning a Unified Representation of Language, Images, and Point Clouds for 3D Understanding443
ULIP is introduced to learn a unified representation of image, text, and 3D point cloud by pre-training with object triplets from the three modalities, using a pre-trained vision-language model that has already learned a common visual and textual space by training with massive image-text pairs.
- 428
- Multi-Hop Knowledge Graph Reasoning with Reward Shaping418
This work reduces the impact of false negative supervision by adopting a pretrained one-hop embedding model to estimate the reward of unobserved facts and counter the sensitivity to spurious paths of on-policy RL by forcing the agent to explore a diverse set of paths using randomly generated edge masks.
- Neural Text Summarization: A Critical Evaluation413
This work critically evaluate key ingredients of the current research setup: datasets, evaluation metrics, and models, and highlights three primary shortcomings: automatically collected datasets leave the task underconstrained and may contain noise detrimental to training and evaluation.
- BERTology Meets Biology: Interpreting Attention in Protein Language Models402
The inner workings of the Transformer is analyzed and it is shown that attention captures the folding structure of proteins, connecting amino acids that are far apart in the underlying sequence, but spatially close in the three-dimensional structure.
- BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset394
A novel approach that employs a diffusion transformer to generate semantically rich CLIP image features, in contrast to conventional VAE-based representations is introduced, which yields both higher training efficiency and improved generative quality.
- UnifiedSKG: Unifying and Multi-Tasking Structured Knowledge Grounding with Text-to-Text Language Models380
The UnifiedSKG framework is proposed, which unifies 21 SKG tasks into a text-to-text format, aiming to promote systematic SKG research, instead of being exclusive to a single task, domain, or dataset.
- CoMatch: Semi-supervised Learning with Contrastive Graph Regularization374
CoMatch is a new semi-supervised learning method that unifies dominant approaches and addresses their limitations, and achieves substantial accuracy improvements on the label-scarce CIFAR-10 and STL-10.
- TOD-BERT: Pre-trained Natural Language Understanding for Task-Oriented Dialogue359
The experimental results show that the pre-trained task- oriented dialogue BERT (ToD-BERT) surpasses BERT and other strong baselines in four downstream task-oriented dialogue applications, including intention detection, dialogue state tracking, dialogue act prediction, and response selection.
- ToD-BERT: Pre-trained Natural Language Understanding for Task-Oriented Dialogues359
The experimental results show that the pre-trained task- oriented dialogue BERT (ToD-BERT) surpasses BERT and other strong baselines in four downstream task-oriented dialogue applications, including intention detection, dialogue state tracking, dialogue act prediction, and response selection.
- Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows354
This work introduces Spider 2.0, an evaluation framework comprising 632 real-world text-to-SQL workflow problems derived from enterprise-level database use cases, and shows that while language models have demonstrated remarkable performance in code generation, they require significant improvement in order to achieve adequate performance for real-world enterprise usage.
- CoSQL: A Conversational Text-to-SQL Challenge Towards Cross-Domain Natural Language Interfaces to Databases344
CoSQL is presented, a corpus for building cross-domain, general-purpose database (DB) querying dialogue systems that includes SQL-grounded dialogue state tracking, response generation from query results, and user dialogue act prediction and a set of strong baselines are evaluated.
- QAFactEval: Improved QA-Based Factual Consistency Evaluation for Summarization341
This work proposes an optimized metric, which they call QAFactEval, that leads to a 14% average improvement over previous QA-based metrics on the SummaC factual consistency benchmark, and also outperforms the best-performing entailment-based metric.
- Jointly Modeling Deep Video and Compositional Text to Bridge Vision and Language in a Unified Framework328
The results show the approach outperforms SVM, CRF and CCA baselines in predicting Subject-Verb-Object triplet and natural sentence generation, and is better than CCA in video retrieval and language retrieval tasks.
- 303
- 300
- FeTaQA: Free-form Table Question Answering298
This work introduces FeTaQA, a new dataset with 10K Wikipedia-based table, question, free-form answer, supporting table cells pairs, and provides two benchmark methods for the proposed task: a pipeline method based on semantic parsing-based QA systems and an end-to-end methodBased on large pretrained text generation models, and shows that FeTaZA poses a challenge for both methods.
- RLHF Workflow: From Reward Modeling to Online RLHF292
This technical report presents the workflow of Online Iterative Reinforcement Learning from Human Feedback (RLHF), and shows that supervised fine-tuning and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets.
- Aguvis: Unified Pure Vision Agents for Autonomous GUI Interaction289
Aguvis is introduced, a unified vision-based framework for autonomous GUI agents that directly operates on screen images, standardizes cross-platform interactions and incorporates structured reasoning via inner monologue, marking the first fully autonomous vision-based GUI agent that operates without closed-source models.
- ULIP-2: Towards Scalable Multimodal Pre-Training for 3D Understanding287
ULIP-2 sheds light on a new paradigm for scalable multimodal 3D representation learning without human annotations and shows significant improvements over existing baselines.
- Bridging Textual and Tabular Data for Cross-Domain Text-to-SQL Semantic Parsing274
This work presents BRIDGE, a powerful sequential architecture for modeling dependencies between natural language questions and relational databases in cross-DB semantic parsing that effectively captures the desired cross-modal dependencies and has the potential to generalize to more text-DB related tasks.
- SParC: Cross-Domain Semantic Parsing in Context273
An in-depth analysis of SParC is provided and it is shown that it introduces new challenges compared to existing datasets and requires generalization to unseen domains due to its cross-domain nature and the unseen databases at test time.
- AdaFrame: Adaptive Frame Selection for Fast Video Recognition272
It is qualitatively demonstrate learned frame usage can indicate the difficulty of making classification decisions; easier samples need fewer frames while harder ones require more, both at instance-level within the same class and at class-level among different categories.
- Streaming Hierarchical Video Segmentation262
This work proposes an approximation framework for streaming hierarchical video segmentation motivated by data stream algorithms: each video frame is processed only once and does not change the segmentation of previous frames.
- UniControl: A Unified Diffusion Model for Controllable Visual Generation In the Wild254
This work introduces UniControl, a new generative foundation model that consolidates a wide array of controllable condition-to-image (C2I) tasks within a singular framework, while still allowing for arbitrary language prompts, and enables pixel-level-precise image generation.
- CodeGen2: Lessons for Training LLMs on Programming and Natural Languages250
This study attempts to render the training of LLMs for program synthesis more efficient by unifying four key components: model architectures, learning methods, infill sampling, and, (4) data distributions.
- BOOKSUM: A Collection of Datasets for Long-form Narrative Summarization245
BookSum is introduced, a collection of datasets for long-form narrative summarization that covers source documents from the literature domain, and includes highly abstractive, human written summaries on three levels of granularity of increasing difficulty: paragraph-, chapter-, and book-level.
- DART: Open-Domain Structured Data Record to Text Generation233
The dataset construction framework effectively merged heterogeneous sources from open domain semantic parsing and spoken dialogue systems by utilizing techniques including tree ontology annotation, question-answer pair to declarative sentence conversion, and predicate unification, all with minimum post-editing.
- GIFT-Eval: A Benchmark For General Time Series Forecasting Model Evaluation230
The General Time Series Forecasting Model Evaluation, GIFT-Eval is introduced, a pioneering benchmark aimed at promoting evaluation across diverse datasets and a comprehensive analysis of 17 baselines, which includes statistical models, deep learning models, and foundation models.
- CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis224
A conversational program synthesis approach via large language models, which addresses the challenges of searching over a vast program space and user intent faced in prior approaches, and outperforms OpenAI’s Codex on the HumanEval benchmark.
- 220
- RNG-KBQA: Generation Augmented Iterative Ranking for Knowledge Base Question Answering216
RnG-KBQA is presented, a Rank-and-Generate approach for KBQA, which remedies the coverage issue with a generation model while preserving a strong generalization capability and achieves new state-of-the-art results on GrailQA and WebQSP datasets.
- Global-Locally Self-Attentive Encoder for Dialogue State Tracking209
This paper proposes the Global-Locally Self-Attentive Dialogue State Tracker (GLAD), which learns representations of the user utterance and previous system actions with global-local modules and shows that this significantly improves tracking of rare states.
- The Regretful Agent: Heuristic-Aided Navigation Through Progress Estimation208
This paper proposes to use a progress monitor developed in prior work as a learnable heuristic for search, and proposes two modules incorporated into an end-to-end architecture that significantly outperforms current state-of-the-art methods using greedy action selection.
- 207
- HIVE: Harnessing Human Feedback for Instructional Visual Editing200
A novel framework to harness human feedback for instructional visual editing (HIVE), which collects human feedback on the edited images and learns a reward function to capture the underlying user preferences, and introduces scalable diffusion model fine-tuning methods that can incorporate human preferences based on the estimated reward.
- Structured Scene Memory for Vision-Language Navigation171
This work proposes a crucial architecture, called Structured Scene Memory (SSM), which is compartmentalized enough to accurately memorize the percepts during navigation and serves as a structured scene representation, which captures and disentangles visual and geometric cues in the environment.
- OpenAgents: An Open Platform for Language Agents in the Wild170
OpenAgents enables general users to interact with agent functionalities through a web user interface optimized for swift responses and common failures while offering developers and researchers a seamless deployment experience on local setups, providing a foundation for crafting innovative language agents and facilitating real-world evaluations.
- Efficient and Robust Question Answering from Minimal Context over Documents170
A simple sentence selector is proposed to select the minimal set of sentences to feed into the QA model, and the overall system achieves significant reductions in training and inference times, with accuracy comparable to or better than the state-of-the-art on SQuAD, NewsQA, TriviaQA and SQuad-Open.
- From image parsing to painterly rendering169
This work presents a semantics-driven approach for stroke-based painterly rendering, based on recent image parsing techniques, which benefits from richer meaningful image semantic information, which leads to better simulation of painting techniques of artists using the high-quality brush dictionary.
- Editing-Based SQL Query Generation for Cross-Domain Context-Dependent Questions166
The interaction history is utilized by editing the previous predicted query to improve the generation quality of SQL queries and the benefit of editing compared with the state-of-the-art baselines which generate SQL from scratch is evaluated.
- CTRLsum: Towards Generic Controllable Text Summarization165
CTRLsum is a generic framework to control generated summaries through a set of keywords, comparable or better than strong pretrained systems in a standard, unconstrained summarization setting.
- 165
- Improving Abstraction in Text Summarization163
This work decomposes the decoder into a contextual network that retrieves relevant parts of the source document, and proposes a pretrained language model that incorporates prior knowledge about language generation to improve the level of abstraction of generated summaries.
- FastIF: Scalable Influence Functions for Efficient Model Interpretation and Debugging157
This work presents FastIF, a set of simple modifications to influence functions that significantly improves their run-time, and uses k-Nearest Neighbors to narrow the search space down to a subset of good candidate data points, identify the configurations that best balance the speed-quality trade-off in estimating the inverse Hessian-vector product, and introduces a fast parallel variant.
- 148
- A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce146
This work revisits GRPO from a reinforce-like algorithm perspective and proposes Reinforce-Rej, a minimal extension of policy gradient that filters both entirely incorrect and entirely correct samples, serving as a lightweight yet effective alternative to more complex RL algorithms.
- Retroformer: Retrospective Large Language Agents with Policy Gradient Optimization146
Experimental results demonstrate that the language agents improve over time and that the approach considerably outperforms baselines that do not properly leverage gradients from the environment, and demonstrates that using policy gradient optimization to improve language agents seems promising and can be applied to optimize other models in the agent architecture to enhance agent performances over time.
- 138
- Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts133
Mirai-MoE is introduced, using a single input/output projection layer while delegating the modeling of diverse time series patterns to the sparse mixture of experts (MoE) within Transformers, which reduces reliance on human-defined heuristics and enables automatic token-level specialization.
- Use All The Labels: A Hierarchical Multi-Label Contrastive Learning Framework132
This paper presents a hierarchical multi-label representation learning framework that can leverage all available labels and preserve the hierarchical relationship between classes and introduces novel hierarchy preserving losses, which jointly apply a hierarchical penalty to the contrastive loss, and enforce the hierarchy constraint.
- BLIP-3: A Family of Open Large Multimodal Models130
BLIP-3, an open framework for developing Large Multimodal Models (LMMs), is introduced and the resulting LMMs demonstrate competitive performance among open-source LMMs with similar model sizes, with the ability to comprehend interleaved image-text inputs.
- 130
- A Survey of Frontiers in LLM Reasoning: Inference Scaling, Learning to Reason, and Agentic Systems129
This categorization provides a systematic understanding of the evolving landscape of LLM reasoning, highlighting emerging trends such as the shift from inference-scaling to learning-to-reason, and the transition to agentic workflows.
- AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials128
This work demonstrates that guided replay using web tutorials is a practical and scalable strategy for training advanced GUI agents, paving the way for more capable and autonomous digital assistants.
- Open Vocabulary Object Detection with Pseudo Bounding-Box Labels122
This work proposes a method to automatically generate pseudo bounding-box annotations of diverse objects from large-scale image-caption pairs that leverages the localization ability of pre-trained vision-language models to generate pseudo Bounding- box labels and then directly uses them for training object detectors.
- Reward-Guided Speculative Decoding for Efficient LLM Reasoning120
Reward-Guided Speculative Decoding is introduced, a novel framework aimed at improving the efficiency of inference in large language models (LLMs) by incorporating a controlled bias to prioritize high-reward outputs in contrast to existing speculative decoding methods that enforce strict unbiasedness.
- DocNLI: A Large-scale Dataset for Document-level Natural Language Inference117
It is demonstrated that, even without fine-tuning, a model pretrained on DocNLI shows promising performance on popular sentence-level benchmarks, and generalizes well to out-of-domain NLP tasks that rely on inference at document granularity.
- Improved Online Conformal Prediction via Strongly Adaptive Online Learning115
New online conformal prediction methods are developed that minimize the strongly adaptive regret, which measures the worst-case regret over all intervals of a fixed length, and it is proved that these methods consistently obtain better coverage and smaller prediction sets than existing methods on real-world tasks.
- BOLAA: Benchmarking and Orchestrating LLM-augmented Autonomous Agents114
A comprehensive comparison of LAA in terms of both agent architectures and LLM backbones is provided and a new strategy to orchestrate multiple LAAs such that each labor LAA focuses on one type of action, BOLAA, where a controller manages the communication among multiple agents.
- Learning From Noisy Anchors for One-Stage Object Detection114
This paper proposes to mitigate noise incurred by imperfect label assignment such that the contributions of anchors are dynamically determined by a carefully constructed cleanliness score associated with each anchor, and conducts extensive experiments on COCO.
- Proposal Learning for Semi-Supervised Object Detection112
This paper presents a proposal learning approach to learn proposal features and predictions from both labeled and unlabeled data and builds a general and high-performance semi-supervised object detection framework, which can be easily adapted to proposal-based object detectors with different backbone architectures.
- VD-BERT: A Unified Vision and Dialog Transformer with BERT111
This work proposes VD-BERT, a simple yet effective framework of unified vision-dialog Transformer that leverages the pretrained BERT language models for Visual Dialog tasks, and adapt BERT for the effective fusion of vision and dialog contents via visually grounded training.
- SummEdits: Measuring LLM Ability at Factual Reasoning Through The Lens of Summarization110
A new protocol for inconsistency detection benchmark creation is proposed and implemented in a 10-domain benchmark called S UMM E DITS, which is 20 times more cost-effective per sample than previous benchmarks and highly reproducible, as it estimates inter-annotator agreement at about 0.9.
- GTA1: GUI Test-time Scaling Agent109
A model that improves grounding of the selected action proposals to its corresponding visual elements is proposed, showing that reinforcement learning (RL) facilitates grounding through inherent objective alignments, rewarding successful clicks on interface elements.
- MCP-Universe: Benchmarking Large Language Models with Real-World Model Context Protocol Servers107
MCP-Universe is introduced, the first comprehensive benchmark specifically designed to evaluate LLMs in realistic and hard tasks through interaction with real-world MCP servers, and implements execution-based evaluators, including format evaluators for agent format compliance, static evaluators for time-invariant content matching, and dynamic evaluators that automatically retrieve real-time ground truth for temporally sensitive tasks.
- FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows"107
This work introduces FaithEval, a novel and comprehensive benchmark tailored to evaluate the faithfulness of LLMs in contextual scenarios across three diverse tasks: unanswerable, inconsistent, and counterfactual contexts.
- Discriminative Nearest Neighbor Few-Shot Intent Detection by Transferring Natural Language Inference99
This paper proposes to boost the discriminative ability by transferring a natural language inference (NLI) model, and achieves more stable and accurate in-domain and OOS detection accuracy than RoBERTa-based classifiers and embedding-based nearest neighbor approaches.
- Lemur: Harmonizing Natural Language and Code for Language Agents98
Lemur and Lemur-Chat are introduced, openly accessible language models optimized for both natural language and coding capabilities to serve as the backbone of versatile language agents, demonstrating balanced proficiencies in both domains, unlike existing open-source models that tend to specialize in either.
- WSLLN:Weakly Supervised Natural Language Localization Networks97
Weakly supervised language localization networks (WSLLN) is proposed to detect events in long, untrimmed videos given language queries to relieve the annotation burden by training with only video-sentence pairs without accessing to temporal locations of events.
- BERT is Not an Interlingua and the Bias of Tokenization95
Cananical Correlation Analysis of the internal representations of a pre- trained, multilingual BERT model reveals that the model partitions representations for each language rather than using a common, shared, interlingual space.
- X-InstructBLIP: A Framework for aligning X-Modal instruction-aware representations to LLMs and Emergent Cross-modal Reasoning92
An efficient and effective framework that integrates multiple modalities (images, 3D, audio and video) to a frozen LLM and demonstrates an emergent ability for cross-modal reasoning (2+ modality inputs).
- Taming MAML: Efficient unbiased meta-reinforcement learning90
A surrogate objective function named, Taming MAML (TMAML), is proposed that adds control variates into gradient estimation via automatic differentiation and improves the quality of gradient estimation by reducing variance without introducing bias.
- 87
- 87
- 87
- Are You Sure? Challenging LLMs Leads to Performance Drops in The FlipFlop Experiment85
The FlipFlop experiment illustrates the universality of sycophantic behavior in LLMs and provides a robust framework to analyze model behavior and evaluate future models.
- How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations85
Several mechanisms within the trained transformers are revealed, such as concrete copying behaviors on both the inputs and the representations, linear ICL capability of the upper layers alone, and a post-ICL representation selection mechanism in a harder mixture setting.
- ThinK: Thinner Key Cache by Query-Driven Pruning84
ThinK is proposed, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels, which achieves a reduction in KV cache memory costs by over 20% compared with vanilla KV cache eviction and quantization methods.
- Agent0: Unleashing Self-Evolving Agents from Zero Data via Tool-Integrated Reasoning80
This work introduces Agent0, a fully autonomous framework that evolves high-performing agents without external data through multi-step co-evolution and seamless tool integration, and establishes a self-reinforcing cycle that continuously produces high-quality curricula.
- 80
- 79
- Beyond the Chat: Executable and Verifiable Text-Editing with LLMs79
Inksync is presented, an editing interface that suggests executable edits directly within the document being edited, leading to more accurate and more efficient editing, and improved user experience.
- Random forests for metric learning with implicit pairwise position dependence74
This work adopts a new angle on the metric learning problem and learns a single metric that is able to implicitly adapt its distance function throughout the feature space and is an order of magnitude faster than state of the art multi-metric methods.
- AgentLite: A Lightweight Library for Building and Advancing Task-Oriented LLM Agent System71
AgentLite is a task-oriented framework designed to enhance the ability of agents to break down tasks and facilitate the development of multi-agent systems and introduces multiple practical applications developed with AgentLite to demonstrate its convenience and flexibility.
- 71
- Universal Natural Language Processing with Limited Annotations: Try Few-shot Textual Entailment as a Start70
This work demonstrates that this framework enables a pretrained entailment model to work well on new entailment domains in a few-shot setting, and shows its effectiveness as a unified solver for several downstream NLP tasks such as question answering and coreference resolution when the end-task annotations are limited.
- 69
- 69
- Double-Hard Debias: Tailoring Word Embeddings for Gender Bias Mitigation68
This work proposes a simple but effective technique, Double Hard Debias, which purifies the word embeddings against such corpus regularities prior to inferring and removing the gender subspace, and preserves the distributional semantics of the pre-trained wordembeddings while reducing gender bias to a significantly larger degree than prior approaches.
- 63
- 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.
- Controllable Abstractive Dialogue Summarization with Sketch Supervision60
A model that can automatically determine or control the number of generated summary sentences for a given dialogue by predicting and highlighting different text spans from the source text and a simple strategy to control the granularity of the final summary.
- Learning to Synthesize Data for Semantic Parsing60
This work proposes a generative model which features a (non-neural) PCFG that models the composition of programs, and a BART-based translation model that maps a program to an utterance.
- StartNet: Online Detection of Action Start in Untrimmed Videos60
The experimental results show that StartNet significantly outperforms the state-of-the-art by 15%-30% p-mAP under the offset tolerance of 1-10 seconds on THUMOS’14, and achieves comparable performance on ActivityNet with 10 times smaller time offset.
- A Survey on Post-training of Large Language Models59
This paper presents the first comprehensive survey of PoLMs, systematically tracing their evolution across five core paradigms: Fine-tuning, which enhances task-specific accuracy; Alignment, which ensures ethical coherence and alignment with human preferences; Reasoning, which advances multi-step inference despite challenges in reward design; Efficiency, which optimizes resource utilization amidst increasing complexity; Integration and Adaptation, which extend capabilities across diverse modalities while addressing coherence issues.
- Photon: A Robust Cross-Domain Text-to-SQL System59
PHOTON is presented, a robust, modular, cross-domain NLIDB that can flag natural language input to which a SQL mapping cannot be immediately determined and effectively improves the robustness of text-to-SQL system against untranslatable user input.
- PerfCodeGen: Improving Performance of LLM Generated Code with Execution Feedback58
PerfCodeGen is proposed, a training-free framework that enhances the performance of LLM-generated code by incorporating feedback based on runtime during test case execution into the self-refinement iterations, highlighting a path forward for more robust and reliable AI-driven software development.
- AgentOhana: Design Unified Data and Training Pipeline for Effective Agent Learning58
This paper introduces \textbf{AgentOhana} as a comprehensive solution to address challenges faced by autonomous agents powered by large language models, and presents \textbf{xLAM-v0.1}, a large action model tailored for AI agents, which demonstrates exceptional performance across various benchmarks.
- 57
- Deep neural language modeling enables functional protein generation across families56
It is demonstrated that a deep learning-based language model can generate functional artificial protein sequences across families, akin to generating grammatically and semantically correct natural language sentences on diverse topics.
- Adapt-and-Adjust: Overcoming the Long-Tail Problem of Multilingual Speech Recognition56
The A2 framework overcomes the long-tail problem via three techniques: exploiting a pretrained multilingual language model (mBERT) to improve the performance of low-resource languages; proposing dual adapters consisting of both language-specific and language-agnostic adaptation with minimal additional parameters; and overcoming the class imbalance.
- 56
- 56
- A Multi-Discriminator CycleGAN for Unsupervised Non-Parallel Speech Domain Adaptation56
A novel generative model based on cyclic-consistent generative adversarial network (CycleGAN) for unsupervised non-parallel speech domain adaptation that employs multiple independent discriminators on the power spectrogram, each in charge of different frequency bands.
- 56
- 56
- 55
- Dense Hierarchical Retrieval for Open-domain Question Answering55
Dense Hierarchical Retrieval (DHR) is proposed, a hierarchical framework that can generate accurate dense representations of passages by utilizing both macroscopic semantics in the document and microscopic semantics specific to each passage.
- WOAD: Weakly Supervised Online Action Detection in Untrimmed Videos55
This work proposes WOAD, a weakly supervised framework that can be trained using only video-class labels and obtains the state-of-the-art results in the tasks of both online per-frame action recognition and online detection of action start.
- 55
- Deep Verifier Networks: Verification of Deep Discriminative Models with Deep Generative Models54
This paper proposes a novel framework --- deep verifier networks (DVN) to detect unreliable inputs or predictions of deep discriminative models, using separately trained deep generative models.
- 54
- 54
- What’s New? Summarizing Contributions in Scientific Literature53
A new task of disentangled paper summarization is introduced, which seeks to generate separate summaries for the paper contributions and the context of the work, making it easier to identify the key findings shared in articles.
- MixQG: Neural Question Generation with Mixed Answer Types53
This paper combines 9 question answering datasets with diverse answer types, including yes/no, multiple-choice, extractive, and abstractive answers, to train a single generative model, MixQG, which outperforms existing work in both seen and unseen domains.
- Discern: Discourse-Aware Entailment Reasoning Network for Conversational Machine Reading53
This work proposes Discern, a discourse-aware entailment reasoning network to strengthen the connection and enhance the understanding for both document and dialog, and splits the document into clause-like elementary discourse units using a pre-trained discourse segmentation model.
- Recognizing Car Fluents from Video53
A spatial-temporal And-Or hierarchical model to represent car fluents is formulated under the latent structural SVM framework and outperforms several highly related baseline methods in terms of car fluent recognition and car part localization.
- BLIP3o-NEXT: Next Frontier of Native Image Generation52
BLIP3o-NEXT unifies text-to-image generation and image editing within a single architecture, demonstrating strong image generation and image editing capabilities.
- Language Models are Hidden Reasoners: Unlocking Latent Reasoning Capabilities via Self-Rewarding52
LaTent Reasoning Optimization (LaTRO), a principled framework that formulates reasoning as sampling from a latent distribution and optimizes it via variational approaches, enables LLMs to concurrently improve both their reasoning process and ability to evaluate reasoning quality, without requiring external feedback or reward models.
- 52
- 51
- Moonshot: Towards Controllable Video Generation and Editing with Multimodal Conditions51
Experiments show that with versatile multimodal conditioning mechanisms, Moonshot demonstrates significant improvement on visual quality and temporal consistency compared to existing models, unveiling its potential to serve as a fundamental architecture for controllable video generation.
- Unsupervised Out-of-Domain Detection via Pre-trained Transformers51
This work tackles the problem of detecting out-of-domain samples with only unsupervised in-domain data by utilizing the latent representations of pre-trained transformers and proposing a simple yet effective method to transform features across all layers to construct out- of-domain detectors efficiently.
- CRMArena-Pro: Holistic Assessment of LLM Agents Across Diverse Business Scenarios and Interactions50
A substantial gap between current LLM capabilities and enterprise demands is highlighted, underscoring the need for advancements in multi-turn reasoning, confidentiality adherence, and versatile skill acquisition.
- LLMs as Factual Reasoners: Insights from Existing Benchmarks and Beyond50
A new protocol for inconsistency detection benchmark creation is proposed and implemented in a 10-domain benchmark called SummEdits, which is 20 times more cost-effective per sample than previous benchmarks and highly reproducible, as it estimates inter-annotator agreement at about 0.9.
- Accurate Annotation of Remote Sensing Images via Active Spectral Clustering with Little Expert Knowledge49
This paper introduces a novel active clustering method for the annotation of high-resolution remote sensing images that can achieve high accuracy in the task of remote sensing image annotation with relatively little expert knowledge, thereby greatly lightening the workload burden and reducing the requirements regarding expert knowledge.
- CodeXEmbed: A Generalist Embedding Model Family for Multiligual and Multi-task Code Retrieval48
Experimental results demonstrate that improving retrieval performance significantly enhances end-to-end Retrieval-Augmented Generation performance for code-related tasks, and this work introduces CodeXEmbed, a family of large-scale code embedding models ranging from 400M to 7B parameters.
- L2CEval : Evaluating Language-to-Code Generation Capabilities of Large Language Models47
This work presents L2CEval, a systematic evaluation of the language-to-code generation capabilities of LLMs on 7 tasks across the domain spectrum of semantic parsing, math reasoning, and Python programming, analyzing the factors that potentially affect their performance.
- 47
- Diversity Empowers Intelligence: Integrating Expertise of Software Engineering Agents46
Diversity Empowered Intelligence (Diversity Empowered Intelligence) functions as a meta-module atop existing SWE agent frameworks, managing agent collectives for enhanced problem-solving and contributes to the growing body of research on collaborative AI systems and their potential to solve complex software engineering challenges.
- 44
- JudgeRank: Leveraging Large Language Models for Reasoning-Intensive Reranking43
JudgeRank is introduced, a novel agentic reranker that emulates human cognitive processes when assessing document relevance and generalizes well across LLMs of various sizes while ensembling them yields even more accurate reranking than individual models.
- A Coarse-to-Fine Framework for Resource Efficient Video Recognition43
LiteEval is a coarse-to-fine framework that dynamically allocates computation on a per-video basis, and can be deployed in both online and offline settings, and adaptively determines on the fly when to read in more discriminative yet computationally expensive features.
- Scalable Chain of Thoughts via Elastic Reasoning42
This work proposes Elastic Reasoning, a novel framework for scalable chain of thoughts that explicitly separates reasoning into two phases--thinking and solution--with independently allocated budgets, and demonstrates that it performs robustly under strict budget constraints, while incurring significantly lower training cost than baseline methods.
- An Investigation of Phone-Based Subword Units for End-to-End Speech Recognition42
Experimental results show that phone-based BPEs tend to yield more accurate recognition systems than the character-based counterpart, and further improvement can be obtained with a novel one-pass joint beam search decoder, which efficiently combines phone- and character- based BPE systems.
- Are Pre-trained Transformers Robust in Intent Classification? A Missing Ingredient in Evaluation of Out-of-Scope Intent Detection41
This work points out the importance of in-domain out-of-scope detection in few-shot intent recognition tasks and illustrates the vulnerability of pre-trained Transformer-based models against samples that are in- domain but out- of-scope (ID-OOS).
- 41
- 40
- 40
- Improving End-to-End Speech Recognition with Policy Learning40
It is shown that joint training improves relative performance by 4% to 13% for the end-to-end model as compared to the same model learned through maximum likelihood, and with policy learning is able to directly optimize on the (otherwise non-differentiable) performance metric.
- 37
- ChatGPT's One-year Anniversary: Are Open-Source Large Language Models Catching up?36
On the first anniversary of ChatGPT, an exhaustive overview of this success is provided, surveying all tasks where an open-source LLM has claimed to be on par or better than ChatG PT.
- 36
- 36
- 36
- Unsupervised Paraphrasing with Pretrained Language Models35
This work adopts a transfer learning approach and proposes a training pipeline that enables pre-trained language models to generate high-quality paraphrases in an unsupervised setting and achieves state-of-the-art performance on both the Quora Question Pair and the ParaNMT datasets.
- 34
- 34
- QAConv: Question Answering on Informative Conversations34
A new question answering (QA) dataset that uses conversations as a knowledge source, focusing on informative conversations, including business emails, panel discussions, and work channels, that provides a new training and evaluation testbed to facilitate QA on conversations research.
- GlueGen: Plug and Play Multi-modal Encoders for X-to-image Generation33
GlueGen is proposed, which applies a newly proposed GlueNet model to align features from single-modal or multi-modal encoders with the latent space of an existing T2I model, allowing for flexible and efficient integration of new functionality into existing T2I models and sheds light on X-to-image (X2I) generation.
- 33
- Latent Domains Modeling for Visual Domain Adaptation33
The results show that the proposed model that automatically discovers latent domains in visual datasets outperforms the existing state-of-the-art methods, and also shows the superiority of multiple latent domain discovery.
- Composed Variational Natural Language Generation for Few-shot Intents31
This paper considers an intent as a combination of a domain and an action, and proposes a composed variational natural language generator (CLANG), a transformer-based conditional variational autoencoder that achieves state-of-the-art performances on two real-world intent detection datasets.
- AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration30
A human-in-the-loop ablation across seven intervention modes reveals that precise, targeted collaboration at high-leverage decision points consistently outperforms both full autonomy and exhaustive step-by-step oversight.
- ESPRIT: Explaining Solutions to Physical Reasoning Tasks30
ESRIT is a framework for commonsense reasoning about qualitative physics in natural language that generates interpretable descriptions of physical events using a data-to-text approach and learns to generate explanations of how the physical simulation will causally evolve.
- Marker-less registration based on template tracking for augmented reality30
An efficient markerless registration algorithm is presented for both outdoor and indoor AR system, and a marker-less AR system is described using this algorithm, and the system architecture and working flow are also proposed.
- SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents29
This work focuses on the development of native Autonomous Single-Agent models for DR featuring minimal web crawling and Python tool integration, and focuses on continual reinforcement learning of reasoning-optimized models to further enhance agentic skills while preserving reasoning ability.
- MAS-ZERO: Designing Multi-Agent Systems with Zero Supervision29
MAS-ZERO is introduced, the first self-evolved, inference-time framework for automatic MAS design, which employs meta-level design to iteratively design, critique, and refine MAS configurations tailored to each problem instance, without requiring a validation set.
- 29
- Fantastic Rewards and How to Tame Them: A Case Study on Reward Learning for Task-oriented Dialogue Systems29
This paper introduces two generalized objectives for reward-function learning, inspired by the classical learning-to-rank literature, and utilizes the learned reward function to guide the training of the E2E ToD agent.
- 29
- MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks28
MASOrchestra is proposed, a training-time framework that formulates MAS orchestration as a function-calling reinforcement learning problem with holistic orchestration, generating an entire MAS at once, achieving consistent improvements on public benchmarks including mathematical reasoning, multi-hop QA, and search-based QA, while achieving more than 10x efficiency over strong baselines.
- RGRecSys28
A robustness evaluation toolkit is presented, Robustness Gym for RecSys (RGRecSys), that allows us to quickly and uniformly evaluate the robustness of recommender system models.
- Spectral Active Clustering via Purification of the k-Nearest Neighbor Graph28
This paper presents a spectral active clustering method that actively select pairwise constraints based on a novel notion of node uncertainty rather than pair uncertainty, and evaluated the framework on three datasets to find the proposed algorithm to be superiorly effective.
- ELECRec: Training Sequential Recommenders as Discriminators27
This work proposes to train the sequential recommenders as discriminators rather than generators, instead of predicting the next item, and trains a discriminator to distinguish if a sampled item is a 'real' target item or not.
- Semi-Supervised Nonlinear Distance Metric Learning via Forests of Max-Margin Cluster Hierarchies27
This work proposes a novel nonlinear metric learning method that uses an iterative, hierarchical variant of semi-supervised max-margin clustering to construct a forest of cluster hierarchies, where each individual hierarchy can be interpreted as a weak metric over the data.
- UNIDOC-BENCH: A Unified Benchmark for Document-Centric Multimodal RAG26
UniDoc-Bench is introduced, the first large-scale, realistic benchmark for MM-RAG built from real-world PDF pages across domains and shows that multimodal text-image fusion RAG systems consistently outperform both unimodal and jointly multimodal embedding-based retrieval, indicating that neither text nor images alone are sufficient and that current multimodal embeddings remain inadequate.
- 26
- Marvista: Exploring the Design of a Human-AI Collaborative News Reading Tool26
The design of Marvista is explored—a human-AI collaborative tool that employs a suite of natural language processing models to provide end-to-end support for reading online news articles to support the entire reading process.
- 26
- A High-Quality Multilingual Dataset for Structured Documentation Translation26
This paper builds and evaluates translation models for seven target languages from English, with several different copy mechanisms and an XML-constrained beam search, and provides a detailed human analysis of gaps between the model output and human translations for real-world applications, including suitability for post-editing.
- UserRL: Training Interactive User-Centric Agent via Reinforcement Learning25
This work systematically vary turn-level reward assignment and trajectory-level score calculation to analyze how different formulations affect learning under the GRPO algorithm, and establishes UserRL as a practical pathway for developing robust user-centric agentic models.
- Pseudo Siamese Network for Few-shot Intent Generation25
Experiments show that the proposed Pseudo Siamese Network achieves state-of-the-art performance for the generalized few shot intent detection task.
- 25
- ReGenesis: LLMs can Grow into Reasoning Generalists via Self-Improvement24
ReGenesis self-synthesizes reasoning paths by converting general reasoning guidelines into task-specific ones, generating reasoning structures, and subsequently transforming these structures into reasoning paths, without the need for human-designed task-specific examples used in existing methods.
- Joint Energy-based Model Training for Better Calibrated Natural Language Understanding Models24
This work explores joint energy-based model (EBM) training during the finetuning of pretrained text encoders for natural language understanding (NLU) tasks and shows that EBM training can help the model reach a better calibration that is competitive to strong baselines, with little or no loss in accuracy.
- Improving Limited Labeled Dialogue State Tracking with Self-Supervision24
This paper presents and investigates two self-supervised objectives: preserving latent consistency and modeling conversational behavior, and encourages a DST model to have consistent latent distributions given a perturbed input, making it more robust to an unseen scenario.
- ProVision: Programmatically Scaling Vision-centric Instruction Data for Multimodal Language Models23
A programmatic approach that employs scene graphs as symbolic representations of images and human-written programs to systematically synthesize vision-centric instruction data is presented, which ensures the interpretability and controllability of the data generation process and scales efficiently while maintaining factual accuracy.
- 23
- Neural Abstract Style Transfer for Chinese Traditional Painting23
A Neural Abstract Style Transfer method for Chinese traditional painting that learns to preserve abstraction and other style jointly end-to-end via a novel MXDoG-guided filter and three fully differentiable loss terms.
- 22
- 22
- 22
- Explicit Memory Tracker with Coarse-to-Fine Reasoning for Conversational Machine Reading22
A new framework of conversational machine reading that comprises a novel Explicit Memory Tracker (EMT) to track whether conditions listed in the rule text have already been satisfied to make a decision and is more interpretable by visualizing the entailment-oriented reasoning process as the conversation flows.
- Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics21
Enterprise Deep Research is presented, a multi-agent system that integrates a Master Planning Agent for adaptive query decomposition, four specialized search agents, an extensible MCP-based tool ecosystem supporting NL2SQL, file analysis, and enterprise workflows, and a Visualization Agent for data-driven insights.
- BOLT: Bootstrap Long Chain-of-Thought in Language Models without Distillation21
A novel approach to enable LLM's LongCoT capacity without distillation from o1-like models or expensive human annotations is introduced, where the authors bootstrap LongCoT (BOLT) from a standard instruct model.
- Empowering Time Series Analysis with Synthetic Data: A Survey and Outlook in the Era of Foundation Models21
This survey provides a comprehensive review of synthetic data for TSFMs and TSLLMs, analyzing data generation strategies, their role in model pretraining, fine-tuning, and evaluation, and identifying future research directions.
- UniTST: Effectively Modeling Inter-Series and Intra-Series Dependencies for Multivariate Time Series Forecasting21
This work proposes a transformer-based model UniTST containing a unified attention mechanism on the flattened patch tokens which reduces the complexity and makes the model feasible for a potentially large number of variates.
- Zero-shot Item-based Recommendation via Multi-task Product Knowledge Graph Pre-Training21
A novel paradigm for the Zero-Shot Item-based Recommendation (ZSIR) task, which pre-trains a model on product knowledge graph (PKG) to refine the item features from PLMs, and discusses how to fine-tune the model on new recommendation task such that the ToA layers are adapted to ZSIR task.
- BatchMixup: Improving Training by Interpolating Hidden States of the Entire Mini-batch21
B ATCH M IXUP can generate new points scattered throughout the space corresponding to the mini-batch and shows superior performance than competitive baselines in improving the performance of NLP tasks while using different ratios of training data.
- Probing Task-Oriented Dialogue Representation from Language Models21
This empirical paper investigates pre-trained language models to find out which model intrinsically carries the most informative representation for task-oriented dialogue tasks and proposes an unsupervised mutual information probe to evaluate the mutual dependence between a real clustering and a representation clustering.
- Unifying Question Answering and Text Classification via Span Extraction.21
A unified, span-extraction approach leads to superior or comparable performance in multi-task learning, low-data and supplementary supervised pretraining experiments on several text classification and question answering benchmarks.
- Improving Tail-Class Representation with Centroid Contrastive Learning20
Interpolative centroid contrastive learning (ICCL) is proposed to improve long-tailed representation learning and shows a significant accuracy gain of 2.8% on the iNaturalist 2018 dataset with a real-world long-tails distribution.
- Lower Bounds for Learning in Revealing POMDPs20
Strong PAC and regret lower bounds for learning in revealing POMDPs are established, and the hard instance construction adapts techniques in distribution testing, which is new to the RL literature and may be of independent interest.
- 20
- MobileAIBench: Benchmarking LLMs and LMMs for On-Device Use Cases19
This work introduces MobileAIBench, a comprehensive benchmarking framework for evaluating mobile-optimized LLMs and LMMs, and provides insights into the performance and feasibility of deploying LLMs and LMMs on mobile platforms.
- MACE: An Efficient Model-Agnostic Framework for Counterfactual Explanation19
A novel framework of Model-Agnostic Counterfactual Explanation (MACE), which adopts a newly designed pipeline that can efficiently handle non-differentiable machine-learning models on a large number of feature values and a novel RL-based method for finding good counterfactual examples and a gradient-less descent method for improving proximity.
- 19
- Agentic Confidence Calibration18
This work introduces the problem of Agentic Confidence Calibration and proposes Holistic Trajectory Calibration (HTC), a novel diagnostic framework that extracts rich process-level features ranging from macro dynamics to micro stability across an agent's entire trajectory, providing a framework for diagnosing and enhancing the reliability of AI agents.
- Agentic Uncertainty Quantification18
A unified Dual-Process Agentic UQ (AUQ) framework that transforms verbalized uncertainty into active, bi-directional control signals and enables the agent to balance efficient execution and deep deliberation dynamically.
- LoCoBench-Agent: An Interactive Benchmark for LLM Agents in Long-Context Software Engineering18
LoCoBench-Agent is introduced, a comprehensive evaluation framework specifically designed to assess LLM agents in realistic, long-context software engineering workflows and establishes a rigorous foundation for measuring agent capabilities, identifying performance gaps, and advancing autonomous software development at scale.
- 18
- Attentive Student Meets Multi-Task Teacher: Improved Knowledge Distillation for Pretrained Models.18
This paper distill the BERT model refined by multi-task learning on seven datasets of the GLUE benchmark into a bidirectional LSTM with attention mechanism, and provides a general learning framework.
- Promptomatix: An Automatic Prompt Optimization Framework for Large Language Models17
This work introduces Promptomatix, an automatic prompt optimization framework that transforms natural language task descriptions into high-quality prompts without requiring manual tuning or domain expertise, and achieves competitive or superior performance compared to existing libraries.
- Automatic Curriculum Expert Iteration for Reliable LLM Reasoning17
Automatic Curriculum Expert Iteration (Auto-CEI) is proposed to enhance LLM reasoning and align responses to the model's capabilities--assertively answering within its limits and declining when tasks exceed them--and compared with various SOTA baselines across logical reasoning, mathematics, and planning tasks, where Auto-CEI achieves superior alignment.
- XGen-7B Technical Report17
This work has trained XGen, a series of 7B parameter models on up to 8K sequence length for up to 1.5T tokens, and finetuned the XGen models on public-domain instructional data, creating their instruction-tuned counterparts (XGen-Inst).
- 17
- Improving Contrastive Learning with Model Augmentation17
This work investigates the possibility of model augmentation to construct view pairs for contrastive SSL and proposes three levels ofmodel augmentation methods: neuron masking, layer dropping, and encoder complementing.
- 17
- 17
- Asynchronous Tool Usage for Real-Time Agents16
This work presents both a conceptual framework and practical tools for creating AI agents capable of fluid, multitasking interactions, and an event-driven finite-state machine architecture for agent execution and prompting, integrated with automatic speech recognition and text-to-speech.
- MathHay: An Automated Benchmark for Long-Context Mathematical Reasoning in LLMs16
This paper introduces MathHay, an automated benchmark designed to assess the long-context mathematical reasoning capabilities of LLMs, and conducts extensive experiments on MathHay to assess the long-context mathematical reasoning abilities of eight top-performing LLMs.
- GReaTer: Gradients over Reasoning Makes Smaller Language Models Strong Prompt Optimizers16
GReaTer, a novel prompt optimization technique that directly incorporates gradient information over task-specific reasoning, enables self-optimization of prompts for open-source, lightweight language models without the need for costly closed-source LLMs.
- 16
- 16
- 16
- 16
- 16
- A Survey of Agent Memory in the Second Half: Towards Self-Evolving and Long-Horizon Agents15
A unified view of foundation agent memory along three dimensions: memory substrate (internal and external), cognitive mechanism (episodic, semantic, sensory, working, and procedural) is provided.
- DyMU: Dynamic Merging and Virtual Unmerging for Efficient VLMs15
DToMe effectively adapts token reduction based on image complexity and, unlike existing systems, provides users more control over computational costs, making it readily applicable to most state-of-the-art VLM architectures.
- 15
- 14
- Self-Abstraction from Grounded Experience for Plan-Guided Policy Refinement14
SAGE is proposed, a framework that enables agents to learn from their own task executions and refine their behavior through self-abstraction, and delivers consistent performance gains across diverse LLM backbones and agent architectures.
- WALT: Web Agents that Learn Tools14
WALT (Web Agents that Learn Tools), a framework that reverse-engineers latent website functionality into reusable invocable tools, achieves higher success with fewer steps and less LLM-dependent reasoning, establishing a robust and generalizable paradigm for browser automation.
- REX: Rapid Exploration and eXploitation for AI Agents14
This paper proposes an enhanced approach for Rapid Exploration and eXploitation for AI Agents called REX, which introduces an additional layer of rewards and integrates concepts similar to Upper Confidence Bound scores, leading to more robust and efficient AI agent performance.
- Model ensemble instead of prompt fusion: a sample-specific knowledge transfer method for few-shot prompt tuning14
This work investigates model ensembles and proposes Sample-specific Ensemble of Source Models (SESoM), a simple ensemble of model predictions based on different source prompts that consistently outperforms the existing models of the same as well as larger parametric scale by a large margin.
- 14
- DIME: An Information-Theoretic Difficulty Measure for AI Datasets14
DIME is an information-theoretic DI for datasets, based on Fano’s inequality and a neural network estimation of the conditional entropy of the sample-label distribution that is well-aligned with empirically observed performance of state-of-the-art machine learning models.
- Adaptive Quantization for Hashing: An Information-Based Approach to Learning Binary Codes14
A novel adaptive quantization (AQ) strategy that adaptively assigns varying numbers of bits to dierent hyperplanes based on their information content is proposed, which provides an information-based schema that preserves the neighborhood structure of data points, and the globally optimal bit-allocation for all hyperplanes.
- CLOVER: A Test Case Generation Benchmark with Coverage, Long-Context, and Verification13
A benchmark to evaluate models' capabilities in generating and completing test cases under specific conditions is presented, and a method to construct retrieval contexts using coverage information is proposed.
- Unanswerability Evaluation for Retrieval Augmented Generation13
UAEval4RAG, a framework designed to evaluate whether RAG systems can handle unanswerable queries effectively, defines a taxonomy with six unanswerable categories, and automatically synthesizes diverse and challenging queries for any given knowledge base with unanswered ratio and acceptable ratio metrics.
- [CASPI] Causal-aware Safe Policy Improvement for Task-oriented Dialogue13
CASPI includes a mechanism to learn fine-grained reward that captures intention behind human response and also offers guarantee on dialogue policy’s performance against a baseline, which demonstrates the effectiveness of this framework on end-to-end dialogue task of the Multiwoz2.0 dataset.
- 13
- GAEA: Graph Augmentation for Equitable Access via Reinforcement Learning13
This work introduces a new class of problems, Graph Augmentation for Equitable Access (GAEA), to enhance equity in networked systems by editing graph edges under budget constraints, and develops a principled, sample- and time- efficient Markov Reward Process (MRP)-based mechanism design framework for GAEA.
- Unsupervised Paraphrase Generation via Dynamic Blocking.13
Dynamic Blocking, a decoding algorithm which enables large-scale pretrained autoregressive models to generate high-quality paraphrases in an unsupervised setting, and a new automatic metric based on self-BLEU and BERTscore which not only discourages the model from copying the input through, but also evaluates text similarity based on distributed representations, hence avoiding reliance on exact keyword matching.
- Seeing is Worse than Believing: Reading People’s Minds Better than Computer-Vision Methods Recognize Actions13
It is indicated that one can read people’s minds better than state-of-the-art computer-vision methods can perform action recognition.
- 13
- 12
- BLIP3-KALE: Knowledge Augmented Large-Scale Dense Captions12
This work introduces BLIP3-KALE, a dataset of 218 million image-text pairs that bridges the gap between descriptive synthetic captions and factual web-scale alt-text and shows the utility of KALE for training more capable and knowledgeable multimodal models.
- Sketch-Fill-A-R: A Persona-Grounded Chit-Chat Generation Framework12
Sketch-Fill-A-R, a framework that uses a persona-memory to generate chit-chat responses in three phases, outperforms a state-of-the-art baseline both quantitatively and qualitatively on the Persona-Chat dataset.
- 12
- 12
- Augmented Cyclic Adversarial Learning for Domain Adaptation.12
This paper proposes an augmented cyclic adversarial learning model that enforces the cycle-consistency constraint through an external task specific model, which encourages the preservation of task-relevant content as opposed to exact reconstruction.
- 12
- Maximum Margin Dirichlet Process Mixtures for Clustering12
This paper proposes a maximum margin Dirichlet process mixture for clustering, which is different from the traditional DPM for parameter modeling, and takes a discriminative clustering approach, by maximizing a conditional likelihood to estimate parameters.
- Convomem Benchmark: Why Your First 150 Conversations Don't Need RAG11
Practical transition points indicate that the small-corpus advantage of conversational memory--where exhaustive search and complete reranking are feasible--deserves dedicated research attention rather than simply applying general RAG solutions to conversation histories.
- MoonShot: Towards Controllable Video Generation and Editing with Motion-Aware Multimodal Conditions11
A new model, MoonShot, conditioning on both image and text for enhanced control, integrating the motion-aware dual cross-attention layer for precise appearance and motion alignment with provided prompts, and the spatiotemporal attention layer for large motion dynamics.
- 11
- 11
- 11
- Efficient Max-Margin Metric Learning11
This paper reformulate the max-margin metric learning problem as a structured support vector machine which can optimize in linear time via a cutting-plane method and proposes a kernelized extention and an approximation method based on matching pursuit that allows linear-time training even in the kernel case.
- AutoResearch AI: Towards AI-Powered Research Automation for Scientific Discovery10
This survey examines the development of AI-powered scientific workflow automation through AutoResearch, defined as the developmental spectrum of AI-powered scientific workflow automation, and proposes five evaluation dimensions--novelty, validity, impact, reliability, and provenance--and shows that AutoResearch autonomy is domain-conditioned.
- 10
- 10
- 10
- Artificial intelligence for streamlined immunofluorescence-based biomarker discovery in prostate cancer.10
The aim is to implement deep learning-based artificial intelligence (AI) models to automate and speed up the analysis of numerous biomarkers and generate prediction models of recurrence and metastasis after surgery.
- 10
- 10
- 10
- 10
- Least-Loaded Expert Parallelism: Load Balancing An Imbalanced Mixture-of-Experts9
Least-Loaded Expert Parallelism (LLEP), a novel EP algorithm that dynamically reroutes excess tokens and associated expert parameters from overloaded devices to underutilized ones, which enables faster and higher-throughput post-training and inference, with ~1.9x faster for gpt-oss-120b.
- LATTE: Learning to Think with Vision Specialists9
By offloading perception to state-of-the-art vision models, this approach enables vision-language models to focus solely on reasoning over high-quality perceptual information.
- 9
- Hierarchical Point Attention for Indoor 3D Object Detection9
This work proposes two novel attention operations as generic hierarchical designs for point-based transformer detectors that builds multi-scale tokens from a single-scale input feature to enable more fine-grained feature learning.
- ToolScan: A Benchmark for Characterizing Errors in Tool-Use LLMs9
SpecTool is introduced, a new benchmark to identify error patterns in LLM output on tool-use tasks and it is shown that even the most prominent LLMs exhibit these error patterns in their outputs.
- Designing and Evaluating Interfaces that Highlight News Coverage Diversity Using Discord Questions9
Three interfaces are designed – the Annotated Article, the Recomposed Article, and the Question Grid – aimed at accompanying news readers in discovering coverage diversity while they read, which reveal coverage diversity and determine each interface’s potential use cases and audiences.
- 9
- Converse: A Tree-Based Modular Task-Oriented Dialogue System9
Converse is a flexible tree-based modular task-oriented dialogue system that uses an and-or tree structure to represent tasks and offers powerful multi-task dialogue management and task dependency and task switching.
- 9
- The Thieves on Sesame Street are Polyglots - Extracting Multilingual Models from Monolingual APIs9
The extraction process extends to local copies initialized from a pre-trained, multilingual model while the victim remains monolingual, and is analyzed to shed light on how such extraction methods succeed.
- 9
- A Unified Framework for Human-Robot Knowledge Transfer.9
A robot system that not only learns in real-time from human demonstrations, but also transfers this knowledge in human-robot interactions is implemented.
- Comprehensive cross-hierarchy cluster agreement evaluation9
The proposed Hierarchy Agreement Index is a novel hierarchy comparison algorithm that holistically compares two hierarchical clustering results and rates their structural agreement on a 0-1 scale and provides more intuitive and reasonable evaluation of the learned hierarchies.
- CHI-Bench: Can AI Agents Automate End-to-End, Long-Horizon, Policy-Rich Healthcare Workflows?8
This work introduces $\chi$-Bench, a benchmark of long-horizon healthcare workflows across three domains: provider prior authorization, payer utilization management, and care management, and raises the hypothesis that similar gaps are likely to surface in other policy-dense, role-composed, irreversible enterprise domains.
- The Illusion of Certainty: Decoupling Capability and Calibration in On-Policy Distillation8
A calibration-aware OPD framework is proposed that estimates empirical confidence from model rollouts, replaces self-reported confidence with this student-grounded target, and distills the revised response through the same self-distillation pipeline, showing that CaOPD achieves Pareto-optimal calibration while maintaining competitive capability.
- Reward Models Identify Consistency, Not Causality8
The analysis reveals that state-of-the-art reward models prioritize structural consistency over causal correctness, suggesting a fundamental limitation in existing reward modeling approaches, emphasizing the need for a shift toward causality-aware reward models that go beyond consistency-driven evaluation.
- Robotic VLA Benefits from Joint Learning with Motion Image Diffusion8
Joint learning with motion image diffusion is proposed, a novel strategy that enhances VLA models with motion reasoning capabilities by extending the VLA architecture with a dual-head design and validating its effectiveness in enhancing the motion reasoning capability of large-scale VLAs.
- Foundational Automatic Evaluators: Scaling Multi-Task Generative Evaluator Training for Reasoning-Centric Domains8
This work trains Foundational Automatic Reasoning Evaluators (FARE), a family of 8B and 20B parameter evaluators, with a simple iterative rejection-sampling supervised finetuning (SFT) approach, and sets the new standard for open-source evaluators, surpassing specialized 70B+ evaluators.
- 8
- Sample-Efficient Learning of POMDPs with Multiple Observations In Hindsight8
This paper proposes an enhanced feedback model called ``multiple observations in hindsight'', where after each episode of interaction with the POMDP, the learner may collect multiple additional observations emitted from the encountered latent states, but may not observe the latent states themselves.
- Structure Extraction in Task-Oriented Dialogues with Slot Clustering8
Empirical results show that the proposed approach outperforms unsupervised baseline models by far in dialogue structure extraction and data augmentation based on extracted structures enriches the surface formats of training data and can achieve a significant performance boost in dialogue response generation.
- On the Diversity and Explainability of Recommender Systems8
A novel framework for improving aggregate recommendation diversity and generating recommendation explanations, which supports a wide variety of models for improving recommendation accuracy and diversity and can also apply to other generic recommender systems for improving Diversity and generating explanations.
- 8
- 8
- Coaction discovery: segmentation of common actions across multiple videos8
The key idea of the approach is to compute an action proposal map for each video based jointly on dynamic object-motion and static appearance semantics, and unsupervisedly cluster each video into atomic action clips, called actoms.
- Fractured Chain-of-Thought Reasoning7
Fractured Sampling is introduced, a unified inference-time strategy that interpolates between full CoT and solution-only sampling along three orthogonal axes: the number of reasoning trajectories, the number of final solutions per trajectory, and the depth at which reasoning traces are truncated.
- SSR: Socratic Self-Refine for Large Language Model Reasoning7
This paper proposes Socratic Self-Refine (SSR), a novel framework for fine-grained evaluation and precise refinement of LLM reasoning that decomposes model responses into verifiable (sub-question, sub-answer) pairs, enabling step-level confidence estimation through controlled re-solving and self-consistency checks.
- Unlocking Anticipatory Text Generation: A Constrained Approach for Large Language Models Decoding7
This work proposes formalizing text generation as a future-constrained generation problem to minimize undesirable behaviors and enforce faithfulness to instructions and demonstrates the effectiveness of the proposed approach across three distinct text generation tasks.
- 7
- SCRIPT: Self-Critic PreTraining of Transformers7
Self-CRItic Pretraining Transformers (SCRIPT) is introduced for representation learning of text and improved sample-efficiency in pretraining and enhanced representations evidenced by improved downstream task performance on GLUE and SQuAD over strong baselines.
- 7
- 7
- 7
- EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environments6
EvoArena is introduced, a benchmark suite that models environment changes as sequences of progressive updates across terminal, software, and social domains, and EvoMem is proposed, a patch-based memory paradigm that records memory evolution as structured update histories, enabling agents to reason about environmental evolution through changes in their memory.
- VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation6
VLAA-GUI, a modular GUI agentic framework built around three integrated components that guide the system on when to Stop, Recover, and Search, achieves top performance on two benchmarks with Linux and Windows tasks.
- Strefer: Empowering Video LLMs with Space-Time Referring and Reasoning via Synthetic Instruction Data6
Strefer is introduced, a synthetic instruction data generation framework designed to equip Video LLMs with spatiotemporal referring and reasoning capabilities, and experimental evaluations show that models trained with data produced by Strefer outperform baselines on tasks requiring spatial and temporal disambiguation.
- 6
- 6
- 6
- 6
- 6
- Salesforce CausalAI Library: A Fast and Scalable Framework for Causal Analysis of Time Series and Tabular Data6
The Salesforce CausalAI Library is introduced, an open-source library for causal analysis using observational data that supports causal discovery and causal inference for tabular and time series data, of discrete, continuous and heterogeneous types.
- 6
- Generating Negative Samples for Sequential Recommendation6
Generate Negative Samples for SR (GenNi) is proposed, where a negative item is sampled at each time step based on the current SR model's learned user preferences toward items, making it scalable to large-scale recommendation tasks.
- 6
- 6
- Online Active Constraint Selection For Semi-Supervised Clustering6
This work presents an active clustering framework which selects pairwise constraints online as clustering proceeds, and proposes an online constraint selection method that actively selects Pairwise constraints by identifying uncertain nodes in the data.
- Trust but Verify: Programmatic VLM Evaluation in the Wild5
This work proposes Programmatic VLM Evaluation (PROVE), a new benchmarking paradigm for evaluating VLM responses to open-ended queries that serves as a foundation for developing nextgeneration VLMs that balance helpfulness with truthfulness.
- SweRank+: Multilingual, Multi-Turn Code Ranking for Software Issue Localization5
SweRank+, a framework that couples SweRankMulti, a cross-lingual code ranking tool, with SweRankAgent, an agentic search setup, for iterative, multi-turn reasoning over the code repository, improves localization over single-pass ranking.
- 5
- ViUniT: Visual Unit Tests for More Robust Visual Programming5
This work proposes Visual Unit Testing (ViUniT), a framework to improve the reliability of visual programs by automatically generating unit tests and introduces four applications of visual unit tests: best program selection, answer refusal, re-prompting, and unsupervised reward formulations for reinforcement learning.
- 5
- CoDA: Coding LM via Diffusion Adaptation5
This release introduces CoDA, a 1.7B-parameter diffusion coder trained on TPU with a fully open-source training pipeline, enabling confidence-guided sampling that keeps inference latency competitive.
- XForecast: Evaluating Natural Language Explanations for Time Series Forecasting5
Two new performance metrics based on simulatability are introduced, assessing how well a human surrogate can predict model forecasts using the explanations, and it is found that numerical reasoning, rather than model size, is the main factor influencing explanation quality.
- 5
- Robustness Evaluation of Transformer-Based Form Field Extractors via Form Attacks5
A novel framework to evaluate the robustness of transformer-based form field extraction methods via form attacks from both OCR level and form level, including OCR location/order rearrangement, form background manipulation and form field-value augmentation is proposed.
- 5
- Field Extraction from Forms with Unlabeled Data5
A novel framework to conduct field extraction from forms with unlabeled data by using the supervisory signal from the pseudo-labels to extract a discriminative token representation from a transformer-based model by modeling the interaction between text in the form.
- 5
- Few-Shot Intent Classification by Gauging Entailment Relationship Between Utterance and Semantic Label5
This work proposes to simplify the NLI-style classification pipeline to be the entailment prediction on the utterance-semantic-label-pair (USLP), and finds that longer and semantically meaningful labels tend to benefit model performance, though the benefit shrinks as more training data is available.
- 5
- Simple Data Augmentation with the Mask Token Improves Domain Adaptation for Dialog Act Tagging5
MaskAugment, a controllable mechanism that augments text input by leveraging the pre-trained Mask token from BERT model is proposed, and an unsupervised teacher-student learning scheme is introduced to examine the domain adaptation of DA taggers.
- 5
- Recent Progress in Deep Reinforcement Learning for Computer Vision and NLP5
This talk begins with general introduction of deep reinforcement learning algorithms, including policy optimization, deep Qlearning, then it highlights the progresses that have achieved in Vision and NLP via DRL.
- 5
- Building Enterprise Realtime Voice Agents from Scratch: A Technical Tutorial4
A complete voice agent is built using Deepgram (streaming STT), vLLM-served LLMs with function calling (streaming text generation), and ElevenLabs (streaming TTS), achieving a measured time-to-first-audio of 755ms (best case 729ms) with full function calling support.
- 4
- 4
- ToolLibGen: Scalable Automatic Tool Creation and Aggregation for LLM Reasoning4
Experimental results demonstrate that the proposed systematic approach to automatically refactor an unstructured collection of tools into a structured tool library significantly improves tool retrieval accuracy and overall reasoning performance across multiple reasoning tasks.
- 4
- 4
- 4
- ARM: Alignment with Residual Energy-Based Model4
Alignment with Residual Energy-Based Model (ARM) is proposed, as a simple and flexible alternative to RLHF methods, that can learn an aligned policy by minimizing a forward Kullback–Leibler divergence from a target policy to a parameteric policy (LLM), instead of a reverse KL as in RLHF methods.
- Bridging the Data Provenance Gap Across Text, Speech and Video4
The largest and first-of-its-kind longitudinal audit across modalities--popular text, speech, and video datasets--from their detailed sourcing trends and use restrictions to their geographical and linguistic representation, finding that multimodal machine learning applications have overwhelmingly turned to web-crawled, synthetic, and social media platforms, such as YouTube, for their training sets.
- 4
- 4
- 4
- 4
- 4
- VoiceAgentRAG: Solving the RAG Latency Bottleneck in Real-Time Voice Agents Using Dual-Agent Architectures3
VoiceAgentRAG is presented, an open-source dual-agent memory router that decouples retrieval from response generation and pre-fetches relevant document chunks into a FAISS-backed semantic cache.
- LZ Penalty: An information-theoretic repetition penalty for autoregressive language models3
The LZ penalty is introduced, a penalty specialized for reducing degenerate repetitions in autoregressive language models without loss of capability and without instances of degenerate repetition that enables open-source reasoning models to operate with greedy decoding without loss of capability.
- Entropy-Based Block Pruning for Efficient Large Language Models3
This work investigates redundancy within Transformer-based models and proposes an entropy-based pruning strategy to enhance efficiency while maintaining performance, and demonstrates that this approach surpasses cosine similarity-based methods in reducing model size while preserving accuracy.
- 3
- 3
- Modeling Uncertainty and Using Post-fusion as Fallback Improves Retrieval Augmented Generation with LLMs3
Four alternative strategies for integrating the retrieved passages with the LLMs are explored, including two single-round methods that utilize chain-of-thought reasoning and two multi-round strategies that incorporate feedback loops that outperform traditional concatenation approach.
- Editing Arbitrary Propositions in LLMs without Subject Labels3
This paper introduces a new dataset, Factual Accuracy Classification Test (FACT), which includes non-binary propositions and for which subject labels are not generally applicable, and therefore is beyond the scope of existing L\&E methods, and shows that with the method editing is possible on FACT.
- 3
- SharPT: Shared Latent Space Prompt Tuning3
The authors' approach outperforms prior soft prompt methods by a significant margin on a variety of tasks such as NLI, sentence completion, QA, conference resolution, word sense disambiguation, and achieves competitive performance compared to finetuning the full model.
- 3
- 3
- 3
- 3
- 3
- Towards the ImageNet-CNN of NLP: Pretraining Sentence Encoders with Machine Translation3
This paper uses the encoder of an attentional sequence-to-sequence model trained for machine translation to initialize models for other, different language tasks, and shows that this transfer improves performance over just using word vectors on a wide variety of common NLP tasks.
- Whisper-AuT: Domain-Adapted Audio Encoder for Efficient Audio-LLM Training2
Whisper-AuT is designed as a drop-in replacement for Whisper in audio-LLM architectures, with the goal of reducing downstream training cost by providing stronger initial audio representations for non-speech domains.
- 2
- 2
- 2
- GeoGNN: Quantifying and Mitigating Semantic Drift in Text-Attributed Graphs2
This work introduces a local PCA--based metric that measures the degree of semantic drift and provides the first quantitative framework to analyze how different aggregation mechanisms affect manifold structure, proposing Geodesic Aggregation, a manifold--aware mechanism that aggregates neighbor information along geodesics via log--exp mappings on the unit sphere.
- 2
- Personalized Multi-task Training for Recommender System2
This paper introduces a novel approach named PMTRec, the first personalized multi-task learning algorithm to obtain comprehensive user/item embeddings from various information sources, and demonstrates its effectiveness in achieving enhanced recommendation accuracy by leveraging multiple tasks simultaneously.
- StructTest: Benchmarking LLMs' Reasoning through Compositional Structured Outputs2
It is demonstrated that StructTest remains challenging even for top-performing models like Deepseek-V3/R1 and GPT-4o, establishing it as a robust proxy for measuring reasoning capabilities.
- 2
- 2
- 2
- Causal Layering via Conditional Entropy2
This paper provides ways to recover layerings of a graph by accessing the data via a conditional entropy oracle, when distributions are discrete.
- 2
- 2
- 2
- 2
- Assessing Local Generalization Capability in Deep Models.2
It is proved that model generalization ability is related to the Hessian, the higher-order “smoothness” terms characterized by the Lipschitz constant of the Hessians, and the scales of the parameters.
- 2
- 2
- Uncertainty reduction for active image clustering via a hybrid global-local uncertainty model2
This work iteratively compute the derivative of the eigenvectors produced by spectral decomposition with respect to each item/image, and combines this with local label entropy provided by the current clustering results in order to estimate the uncertainty reduction potential of each item in the dataset.
- Prompt Optimization Via Diffusion Language Models1
This work proposes a diffusion-based framework for prompt optimization that leverages Diffusion Language Models to iteratively refine system prompts through masked denoising and shows that moderate diffusion step counts provide the best balance between refinement quality and stability.
- Enterprise Sales Copilot: Enabling Real-Time AI Support with Automatic Information Retrieval in Live Sales Calls1
Real-time AI-powered assistant SalesCopilot automatically detects customer questions, retrieving relevant information from the product database, and displaying concise answers on the representative's dashboard in seconds, representing a 14xspeedup compared to manual CRM search in an internal study.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Introduction to the Special Issue on Knowledge Transferring for Recommender Systems1
The role of knowledge transfer in advancing recommender systems is explored and featured works address a variety of challenges, including cold-start problems, cross-domain recommendations, knowledge graph integration, federated learning, explainability, and so on.
- 1
- 1
- 1
- Learning to Play General-Sum Games against Multiple Boundedly Rational Agents1
It is shown that no-regret dynamics can identify the worst possible strategic responses of other agents in poly-time in smooth games and is proposed a framework that uses this policy evaluation method for efficiently learning a robust principal policy using RL.
- 1
- 1
- Numerical Correlation in Text1
A new task, Numerical Correlation in Text, which requires models to identify the correlation between two numbers in a sentence is proposed and a new dataset is introduced, which contains over 2,000 Wikipedia sentences with two numbers and their correlation labels.
- 1
- BigIssue: A Realistic Bug Localization Benchmark1
The introduction of BigIssue is proposed, a general benchmark for realistic bug localization and a motivation to improve bug localization capabilities of models through attention to the full repository context, to advance the state of the art in bug localization.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Style Recognition and Kinship Understanding1
This chapter addresses two novel applications using deep learning: style recognition and kinship understanding and proposes a new familial feature extraction method towards better kinship parsing.
- Entropy Penalty: Towards Generalization Beyond the IID Assumption1
This work proposes a regularization approach based on IB, called Entropy Penalty, that reduces the model's dependence on spurious features-- features corresponding to such spurious correlations, and shows that it is able to generalize well on vanilla MNIST, MNIST-M and SVHN datasets.
- 1
- MP28-01 ARTIFICIAL INTELLIGENCE (AI) ACCURATELY AUTOMATE AND SPEED IMMUNOFLUORESCENCE (IF)-BASED DISCOVERY AND VALIDATION OF NOVEL PROGNOSTIC AND PREDICTIVE BIOMARKERS IN PROSTATE CANCER1
An AI model was implemented to automate the analysis of biomarkers by recognizing specific expression patterns of the markers of interest in epithelial cells and normal stromal tissue to translate the finding into predictions of recurrence and metastasis after radical prostatectomy.
- 1
- 1
- –
- –
- –
- Position: Vector Prompt Interfaces Should Be Exposed to Enable Customization of Large Language Models–
This position paper argues that model providers should expose vector prompt inputs as part of the public interface for customizing LLMs, and supports this position with diagnostic evidence showing that vector prompt tuning continues to improve with increasing supervision whereas text-based prompt optimization saturates early.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Enabling High Data Throughput Reinforcement Learning on GPUs: A Domain Agnostic Framework for Data-Driven Scientific Research–
WarpSci is introduced, a domain agnostic framework designed to overcome crucial system bottlenecks encountered in the application of reinforcement learning to intricate environments with vast datasets featuring high-dimensional observation or action spaces, and eliminates the need for data transfer between CPU and GPU.
- –
- –
- –
- –
- –
- –
- –
- Guided Adaptive Credit Assignment for Sample Efficient Policy Optimization–
Empirical studies show that the proposed guided adaptive credit assignment method improves the sample efficiency of the state-of-the-art policy optimization approaches.
- –
- –
- –
- –
- –
- Towards a parts-based approach to sub-cortical brain structure parsing–
This work proposes a method that defines a principled parts-based representation of the sub-cortical brain structures based on the pictorial structures model and jointly models the appearance of each part as well as the layout of the parts as a whole.
- Object category recognition using boosting tree with heterogenous features–
The proposed approach integrates the Adaboost algorithm into the decision tree structure, called DB-Tree, and each tree node combines a number of weak classifiers into a strong classifier (a conditional posterior probability), which can output the overall posterior probability.
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.