3/10/2026
Explainable Token-level Noise Filtering for LLM Fine-tuning Datasets
Token-level noise filtering method for LLM fine-tuning datasets, addressing mismatch between sentence-level annotation and token optimization.
3/10/2026
LongAudio-RAG: Event-Grounded Question Answering over Multi-Hour Long Audio
LongAudio-RAG hybrid framework for QA over multi-hour audio with temporal grounding and minimal hallucination.
3/10/2026
Understand Then Memory: A Cognitive Gist-Driven RAG Framework with Global Semantic Diffusion
CogitoRAG framework simulates human cognitive memory for retrieval-augmented generation, using semantic diffusion to preserve integrity.
3/10/2026
Condition-Gated Reasoning for Context-Dependent Biomedical Question Answering
Condition-gated reasoning system for biomedical QA that handles patient-specific conditional logic in clinical decision-making.
3/10/2026
Conformal Tradeoffs: Guarantees Beyond Coverage
Analyzes deployment tradeoffs in conformal predictors beyond coverage, examining commit vs defer vs error exposure operational metrics.
3/10/2026
CrystaL: Spontaneous Emergence of Visual Latents in MLLMs
CrystaL enables latent chain-of-thought reasoning in multimodal LLMs without predefined supervision, improving vision-language integration.
3/10/2026
MrBERT: Modern Multilingual Encoders via Vocabulary, Domain, and Dimensional Adaptation
ModernBERT-based multilingual encoder family (150M-300M params) pretrained on 35 languages with domain and dimensional adaptation.
3/10/2026
UniWhisper: Efficient Continual Multi-task Training for Robust Universal Audio Representation
Continual multi-task training framework for universal audio representation across speech, environmental sounds, and music.
3/10/2026
CeRA: Breaking the Linear Ceiling of Low-Rank Adaptation via Manifold Expansion
CeRA improves low-rank adaptation for LLM fine-tuning by adding manifold expansion via gating and dropout, addressing linear limitations.
3/10/2026
Scaling Search Relevance: Augmenting App Store Ranking with LLM-Generated Judgments
Combines behavioral and textual relevance signals using LLMs to improve app store search ranking at scale.
3/10/2026
Attn-QAT: 4-Bit Attention With Quantization-Aware Training
First systematic 4-bit quantization-aware training study for attention mechanisms enabling end-to-end FP4 computation on emerging GPUs.
3/10/2026
PEPA: a Persistently Autonomous Embodied Agent with Personalities
PEPA: embodied AI agent framework with personality-driven persistent autonomy enabling self-sustaining goals without external task specification.
3/10/2026
Conformal Prediction for Risk-Controlled Medical Entity Extraction Across Clinical Domains
Conformal prediction framework providing finite-sample coverage guarantees for LLM-based medical entity extraction across clinical domains.
3/10/2026
Human-Certified Module Repositories for the AI Age
Architectural model for trustworthy AI-assisted software via human-certified module repositories ensuring reliability of AI-assembled systems.
3/10/2026
iGVLM: Dynamic Instruction-Guided Vision Encoding for Question-Aware Multimodal Understanding
iGVLM: framework enabling dynamic instruction-guided vision encoding in LVLMs for task-specific visual understanding.
3/10/2026
Test-Time Meta-Adaptation with Self-Synthesis
MASS: meta-learning framework enabling LLMs to self-adapt at test time by generating synthetic training data for improved downstream performance.
3/10/2026
Neuro-Symbolic Financial Reasoning via Deterministic Fact Ledgers and Adversarial Low-Latency Hallucination Detector
Neuro-symbolic approach combining LLMs with deterministic fact ledgers and hallucination detection for financial reasoning without arithmetic errors.
3/10/2026
Bridging Domains through Subspace-Aware Model Merging
Research on merging task-specific models into consolidated ones, analyzing parameter competition and domain generalization effects.
3/10/2026
vLLM Hook v0: A Plug-in for Programming Model Internals on vLLM
vLLM Hook v0 plugin enabling programmable access to LLM model internals for test-time alignment and inference optimization in vLLM serving engine.
3/10/2026
How Attention Sinks Emerge in Large Language Models: An Interpretability Perspective
Interpretability study on attention sinks in LLMs, explaining why models allocate disproportionate attention to specific tokens including first token bias.
3/10/2026
FuzzingRL: Reinforcement Fuzz-Testing for Revealing VLM Failures
FuzzingRL approach using reinforcement learning for fuzz testing Vision Language Models to automatically generate failure-inducing queries.
3/10/2026
Switchable Activation Networks
Switchable Activation Networks that dynamically select activation functions for computational efficiency in LLMs and vision-action models during inference.
3/10/2026
Know When You're Wrong: Aligning Confidence with Correctness for LLM Error Detection
Method to align LLM confidence scores with correctness using output token probabilities for reliable error detection and hallucination identification.
3/10/2026
LegoNet: Memory Footprint Reduction Through Block Weight Clustering
LegoNet compression technique for neural networks using block weight clustering to reduce memory footprint for embedded device deployment.
3/10/2026
CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training
CapTrack benchmark for evaluating multi-faceted forgetting in LLM post-training beyond parametric knowledge loss, addressing domain adaptation challenges.
3/10/2026
Consensus is Not Verification: Why Crowd Wisdom Strategies Fail for LLM Truthfulness
Research showing majority-voting and ensemble inference methods fail to improve LLM truthfulness without external verification, unlike in math/code domains.
3/10/2026
OptiRoulette Optimizer: A New Stochastic Meta-Optimizer for up to 5.3x Faster Convergence
OptiRoulette meta-optimizer that dynamically selects update rules during training via warmup locking and random sampling. Torch-compatible drop-in component achieving 5.3x faster convergence.
3/10/2026
RACER: Risk-Aware Calibrated Efficient Routing for Large Language Models
RACER system for efficient multi-model LLM routing formulated as risk-aware optimization problem. Extends base routers to minimize cost-performance trade-off.
3/10/2026
Evo: Autoregressive-Diffusion Large Language Models with Evolving Balance
Novel language model combining autoregressive and diffusion-based generation through latent trajectory modeling with evolving balance parameter.
3/10/2026
Not all tokens are needed(NAT): token efficient reinforcement learning
Framework for token-efficient reinforcement learning in LLMs. Proposes NAT to reduce computational cost of backpropagation over long chain-of-thought trajectories during training.
3/10/2026
Reward Under Attack: Analyzing the Robustness and Hackability of Process Reward Models
Research on vulnerabilities in Process Reward Models used in LLM reasoning pipelines. Introduces diagnostic framework to quantify adversarial exploitability and fluency-logic dissociation.
3/10/2026
From ARIMA to Attention: Power Load Forecasting Using Temporal Deep Learning
Empirical comparison of ARIMA, LSTM, BiLSTM, and Transformer for short-term power load forecasting.
3/10/2026
Advances in GRPO for Generation Models: A Survey
Survey of Group Relative Policy Optimization for aligning generative models with human preferences.
3/10/2026
Grouter: Decoupling Routing from Representation for Accelerated MoE Training
Grouter: decoupled routing method for accelerating Mixture-of-Experts training with structural priors.
3/10/2026
Leakage Safe Graph Features for Interpretable Fraud Detection in Temporal Transaction Networks
Leakage-safe graph feature extraction for fraud detection in temporal transaction networks.
3/10/2026
Graph Property Inference in Small Language Models: Effects of Representation and Inference Strategy
Graph property inference in small language models: effects of representation on structured reasoning.
3/10/2026
SmartBench: Evaluating LLMs in Smart Homes with Anomalous Device States and Behavioral Contexts
SmartBench: evaluation benchmark for LLMs in smart home environments with anomalous device detection.
3/10/2026
HEARTS: Benchmarking LLM Reasoning on Health Time Series
HEARTS: benchmark for evaluating LLM reasoning on diverse health time series tasks and modalities.
3/10/2026
SR-TTT: Surprisal-Aware Residual Test-Time Training
SR-TTT: test-time training for LLMs with infinite context via fast weights, improving long-context reasoning.
3/10/2026
Trust Aware Federated Learning for Secure Bone Healing Stage Interpretation in e-Health
Trust-aware federated learning framework for bone healing classification in distributed medical environments.
3/10/2026
ERP-RiskBench: Leakage-Safe Ensemble Learning for Financial Risk
Ensemble learning framework for financial risk detection in ERP systems with leakage-safe evaluation.
3/10/2026
Scaling Agentic Capabilities, Not Context: Efficient Reinforcement Finetuning for Large Toolspaces
ATLAS: reinforcement finetuning framework for scaling agent capabilities with large toolspaces using small language models.
3/10/2026
From Statistical Fidelity to Clinical Consistency: Scalable Generation and Auditing of Synthetic Patient Trajectories
Synthetic EHR generation pipeline with clinical consistency validation for privacy-preserving health data sharing.
3/10/2026
ProtAlign: Contrastive learning paradigm for Sequence and structure alignment
ProtAlign: contrastive learning framework for protein sequence-structure alignment using language models.
3/10/2026
Regression Models Meet Foundation Models: A Hybrid-AI Approach to Practical Electricity Price Forecasting
Hybrid approach combining time series foundation models with regression models for electricity price forecasting capturing temporal and cross-variate patterns.
3/10/2026
Safe Transformer: An Explicit Safety Bit For Interpretable And Controllable Alignment
Safe Transformer: Modular approach adding explicit safety bit as interpretable information bottleneck for controllable alignment in language models.
3/10/2026
Orion: Characterizing and Programming Apple's Neural Engine for LLM Training and Inference
Orion: First open system enabling direct LLM training and inference on Apple Neural Engine (ANE) with compiler pipeline and on-device training support.
3/10/2026
Don't Freeze, Don't Crash: Extending the Safe Operating Range of Neural Navigation in Dense Crowds
Reinforcement learning approach for safe neural navigation in variable-density crowds that generalizes beyond training conditions.
3/10/2026
Rank-Factorized Implicit Neural Bias: Scaling Super-Resolution Transformer with FlashAttention
Super-resolution transformer optimization using FlashAttention with rank-factorized implicit neural bias to reduce computational burden.
3/10/2026
Heterogeneous Decentralized Diffusion Models
Efficient decentralized framework for training diffusion models with heterogeneous objectives across isolated experts with reduced computational requirements.