Topics

Papers deriving tight regret bounds and complexity analysis for learning algorithms, focusing on minimax-optimal convergence rates and dimension-dependent lower bounds.

Papers on evaluating and benchmarking multimodal models' ability to understand, generate, and edit video content by integrating visual, textual, and physical reasoning.

Papers on learning and using world models that predict environment dynamics conditioned on robot or agent actions for planning and embodied task execution.

Papers on segmenting anatomical structures and pathological lesions in medical images (CT, MRI, ultrasound) using deep learning and specialized architectures.

Papers on verifying the trustworthiness and correctness of LLM-based autonomous agents through tracing, benchmarking, and auditing their reasoning and actions in real-world deployments.

Papers using Gaussian splatting and 3D representations to reconstruct scenes, objects, and surfaces from images, depth, or point cloud data.

Papers on training reward models and critics to learn human preferences through reinforcement learning, including approaches for visual reasoning, multi-objective alignment, and preference-based optimization.

Papers examining how large language models represent and process belief states, moral reasoning, and conversational understanding—comparing their cognitive representations to human mental models.

Papers on training models to distill their own knowledge during on-policy learning, where a model learns from its own generated reasoning or behaviors without requiring a separate teacher network.

Papers examining biases, artifacts, and consistency issues when using large language models as evaluators for text quality, ranking decisions, and comparative judgments.

Papers examining how to measure, audit, and computationally understand aesthetic and compositional properties in music, visual design, and artwork through machine perception and vision-language models.

Papers on reducing computational costs in vision-language models by selectively pruning or compressing visual tokens through attention mechanisms and encoder optimization.

Papers on using neural networks to learn and approximate differential operators and PDE solutions by incorporating physical constraints and domain knowledge into the learning process.

Papers on detecting, evaluating, and mitigating security threats in large language models, including prompt injection attacks, skill extraction, and safety failures through red-teaming and fuzzing approaches.

Papers on detecting and mitigating demographic bias in machine learning systems through auditing methods, counterfactual testing, and fairness-aware filtering or adjustments for confounding variables.

Papers addressing traceability, attribution, and responsibility concerns in LLM-assisted workflows including financial analysis, systematic reviews, and citation tracking.

Papers applying foundation models and deep learning to predict molecular properties, chemical reactions, and materials structures, leveraging transfer learning and multi-modal representations.

Papers applying deep learning and ensemble methods to predict weather phenomena like precipitation, temperature, and atmospheric profiles at various timescales.

Papers addressing the communication efficiency, privacy preservation, and distributed coordination challenges in federated learning systems across decentralized clients and wireless networks.

Papers addressing limitations and failure cases of retrieval-augmented generation systems, including hallucinations, misinformation, biases, and methods to improve answer reliability through better retrieval and reasoning strategies.

Papers on using language models with reinforcement learning and symbolic reasoning to automatically prove mathematical theorems and verify formal systems.

Papers addressing security vulnerabilities and defenses in mixture-of-experts models, including backdoor attacks, adversarial attacks, bit-flip attacks, and watermarking for protection and integrity verification.

Papers on reducing memory and computational overhead during inference through KV cache optimization, prefix caching, and efficient decoding strategies for language and protein models.

Papers on integrating multiple physiological signals (ECG, PPG, EHR) and medical imaging modalities to improve clinical predictions under missing data and signal degradation, with emphasis on survival and treatment outcome estimation.

Papers on routing and orchestration strategies for directing tasks across multiple LLM agents in online, multi-step workflows, including adaptive delegation and real-time scheduling.

Papers on detecting and monitoring plant health and environmental stress using aerial imagery, spectral measurements, and temporal sensor data from remote sensing platforms.

Papers on predicting and optimizing traffic flow through multi-agent modeling of vehicle interactions, collective behavior, and autonomous driving coordination in shared corridors and intersections.

Papers on building and evolving reusable skill libraries that enable agents to handle long-horizon tasks through persistent knowledge and skill composition.

Papers on building, evaluating, and improving systems for converting spoken audio to text across diverse languages, domains, and acoustic conditions.

Papers on optical and computational recognition of text in non-standard formats including historical manuscripts, sign language, Braille, and ancient scripts.

Papers on designing and iteratively improving test harnesses and evaluation frameworks that guide agent behavior and performance in coding and operational tasks.

Papers on applying low-rank adaptation methods like LoRA to continual learning scenarios where models must learn new tasks sequentially while preserving learned knowledge through efficient parameter updates.

Papers using diffusion models and equivariant neural networks to solve ill-posed inverse problems in imaging and physical field reconstruction with learned or adaptive regularization.

Papers on reducing model size and computational cost through quantization (including low-bit formats like 4-bit), pruning, and second-order optimization methods like Kronecker-factored Hessians applied to large language models.

Papers applying causal inference methods to optimize recommendations and forecasting in e-commerce and retail, addressing confounding and estimating treatment effects at the product or item level.

Papers on incorporating morphological structure and positional awareness into pre-trained models, particularly for morphologically rich and low-resource languages.

Papers on methods and challenges for detecting machine-generated text, including authorship attribution, robustness to confounds, and spectral or linguistic signatures that distinguish AI-written content from human writing.

Papers on learning unified embedding representations across modalities (image, text, audio) to enable cross-modal retrieval and matching without task-specific training.

Papers investigating how tokenizers handle phonetic, orthographic, and linguistic properties across languages, and how tokenization choices affect cross-lingual alignment and representation learning.

Papers on accelerating and improving inference in diffusion-based language models through techniques like speculative decoding, rejection sampling, and length-aware token generation to mitigate repetition and degeneration.

Papers on detecting objects across multiple views, scales, or sensor modalities using deformable networks and fusion techniques, including challenging domains like underwater, maritime, and panoramic scenarios.

Papers on evaluating and improving the reliability and calibration of vision-language models applied to medical imaging, pathology, and visual question answering tasks.

Papers combining score-based methods with uncertainty quantification and conformal risk control to improve predictions on stochastic dynamical systems and physical fields using unbiased gradient estimation.

Papers on using large language models and AI agents to automate and accelerate hardware design workflows, from chip accelerators to microarchitecture exploration and neuromorphic systems.

Papers on controlling LLM behavior by manipulating learned directions or vectors in activation space without modifying model weights.

Papers on runtime control mechanisms, identity verification, and organizational enforcement frameworks for managing autonomous AI agents in production deployments.

Papers on evaluating and aligning language models against structured reference documents, whether through model cards, benchmark datasets, or rule-based review of domain-specific documentation.

Papers on applying graph neural networks to graphs where nodes have text features or descriptions, addressing how to effectively integrate textual information with graph structure in learning tasks.

Papers evaluating how well language models follow clinical decision guidelines and protocols when making medical diagnoses and treatment recommendations.

Papers on designing and optimizing memory systems for AI agents, focusing on long-term context retention, personalized storage, and query-aware retrieval to support collaborative and multi-modal agent interactions.

Papers on interpreting neural network predictions through human-understandable concepts and counterfactual explanations, with applications to time series and multimodal domains.

Papers on modifying and understanding neural circuits during post-training and fine-tuning through targeted adaptation techniques and behavioral circuit analysis.

Papers creating and evaluating benchmarks specifically designed to assess language models and AI agents on mental health and psychological tasks, including longitudinal monitoring and conversational assessment.

Papers on using large language models as autonomous agents to design, execute, and iterate through scientific workflows including protein folding, control systems, and experimental automation.

Papers combining discrete choice modeling with planning and trajectory prediction, bridging behavioral learning from user preferences or demonstrations with goal-directed decision-making in navigation and navigation tasks.

Papers on segmenting scenes and objects using open-vocabulary models, foundation models, and generative approaches to enable zero-shot or training-free segmentation across diverse visual domains including LiDAR, UAV imagery, and cluttered environments.

Papers applying event-based cameras (neuromorphic sensors that capture asynchronous pixel-level changes) to video understanding tasks like anomaly detection, object discovery, and reflection removal.

Papers on retrieving answers from knowledge graphs through multi-hop reasoning over entities and subgraphs, with focus on integration with retrieval-augmented generation systems.

Papers on enhancing resolution and reconstruction quality of brain imaging modalities (MRI, EEG, CT) while preserving or recovering fine anatomical details like white-matter lesions.

Papers on selectively inserting, editing, or removing specific knowledge from language models while preserving existing capabilities and generalization.