3/17/2026
Low-Dimensional and Transversely Curved Optimization Dynamics in Grokking
Geometric analysis of optimization dynamics in grokking; shows transformers train in low-dimensional subspaces with detailed PCA findings.
3/17/2026
Early-Warning Signals of Grokking via Loss-Landscape Geometry
Research on early-warning signals for grokking phenomenon via loss-landscape geometry analysis across sequence learning benchmarks.
3/17/2026
When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators
Investigation of text-to-image diffusion models' effectiveness as synthetic data generators, revealing performance regression when used for training data generation.
3/17/2026
LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration
LESA method for accelerating diffusion models through learnable stage-aware predictors that adapt to stage-dependent dynamics with feature caching.
3/17/2026
Survey on Neural Routing Solvers
Survey of neural routing solvers that use deep learning to tackle vehicle routing problems by learning implicit heuristic rules from data.
3/17/2026
A Comprehensive Evaluation of LLM Unlearning Robustness under Multi-Turn Interaction
Study of LLM unlearning robustness in multi-turn interactive settings, addressing safety, privacy, and legal concerns in machine unlearning.
3/17/2026
A Gauge Theory of Superposition: Toward a Sheaf-Theoretic Atlas of Neural Representations
Discrete gauge-theoretic framework for understanding superposition in LLMs using sheaf theory and local semantic charts instead of global dictionaries.
3/17/2026
RMBench: Memory-Dependent Robotic Manipulation Benchmark with Insights into Policy Design
arXiv paper introducing RMBench robotic manipulation benchmark emphasizing memory-dependent tasks and policy design insights.
3/17/2026
CHLU: The Causal Hamiltonian Learning Unit as a Symplectic Primitive for Deep Learning
arXiv paper proposing Causal Hamiltonian Learning Unit as deep learning primitive for temporal dynamics addressing LSTM/Neural ODE tradeoffs.
3/17/2026
On Google's SynthID-Text LLM Watermarking System: Theoretical Analysis and Empirical Validation
arXiv paper analyzing Google's SynthID-Text watermarking system for LLM-generated text detection using tournament-based methods.
3/17/2026
Do Mixed-Vendor Multi-Agent LLMs Improve Clinical Diagnosis?
arXiv paper evaluating multi-vendor LLM agent teams for clinical diagnosis, testing whether diversity reduces correlated failure modes.
3/17/2026
Aura: Universal Multi-dimensional Exogenous Integration for Aviation Time Series
arXiv paper on time series forecasting with multi-dimensional exogenous integration for industrial applications like aviation.
3/17/2026
A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness
Study showing LLM-as-a-Judge evaluation frameworks are unreliable for safety assessment, demonstrating vulnerability to distribution shifts in red-teaming.
3/17/2026
HEARTS: Benchmarking LLM Reasoning on Health Time Series
HEARTS benchmark for evaluating LLM reasoning on health time series across multiple physiological modalities and temporal dependencies.
3/17/2026
Failure Detection in Chemical Processes Using Symbolic Machine Learning: A Case Study on Ethylene Oxidation
Symbolic ML approach for failure detection in chemical processes, emphasizing interpretability and safety over neural methods; includes ethylene oxidation case study.
3/17/2026
Agora: Teaching the Skill of Consensus-Finding with AI Personas Grounded in Human Voice
AI-powered platform using LLM personas to teach deliberative democratic skills and consensus-finding through simulated discussion scenarios.
3/17/2026
AgrI Challenge: A Data-Centric AI Competition for Cross-Team Validation in Agricultural Vision
ML competition for agricultural vision focusing on data-centric approaches and model generalization under distribution shifts in real field conditions.
3/17/2026
Distributional Regression with Tabular Foundation Models: Evaluating Probabilistic Predictions via Proper Scoring Rules
Evaluation framework for tabular foundation models using proper scoring rules to assess full predictive distributions, not just point estimates.
3/17/2026
Concept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustness
Fine-tuning method for Vision Transformers using concept guidance to reduce spurious correlations and improve robustness to distribution shifts.
3/17/2026
A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic
Clinical feasibility study of LLM-based conversational diagnostic AI (AMIE) in real primary care workflows with safety evaluation.
3/17/2026
Emotion is Not Just a Label: Latent Emotional Factors in LLM Processing
Study of emotion as latent factor affecting LLM reasoning and attention mechanisms, rather than just a prediction target.
3/17/2026
Open-World Motion Forecasting
Motion forecasting for autonomous vehicles handling open-world scenarios with imperfect perception and evolving object taxonomies.
3/17/2026
EvoDriveVLA: Evolving Autonomous Driving Vision-Language-Action Model via Collaborative Perception-Planning Distillation
Vision-language-action model for autonomous driving using perception-planning distillation to improve visual encoding and trajectory planning stability.
3/17/2026
Evaluating Adjective-Noun Compositionality in LLMs: Functional vs Representational Perspectives
Research on how LLMs handle compositional language tasks (adjective-noun relationships), comparing external performance with internal model representations.
3/17/2026
Evaluation format, not model capability, drives triage failure in the assessment of consumer health AI
Research comparing LLM performance in healthcare triage across evaluation formats. Shows evaluation methodology significantly affects model assessment outcomes.
3/17/2026
KEPo: Knowledge Evolution Poison on Graph-based Retrieval-Augmented Generation
KEPo: research on knowledge graph poisoning attacks against GraphRAG systems. Analyzes vulnerabilities when LLMs rely on external databases.
3/17/2026
RoboClaw: An Agentic Framework for Scalable Long-Horizon Robotic Tasks
RoboClaw: Agentic framework unifying data collection, policy learning, and deployment for long-horizon robotic manipulation using Vision-Language-Action systems.
3/17/2026
Truth as a Compression Artifact in Language Model Training
Controlled experiments showing language models prefer correct answers due to data compressibility structure rather than truth, using small transformers on contradictory corpora.
3/17/2026
MobileKernelBench: Can LLMs Write Efficient Kernels for Mobile Devices?
MobileKernelBench benchmark evaluating LLM capability to generate efficient computational kernels for mobile devices, with systematic investigation of code generation limits.
3/17/2026
LoV3D: Grounding Cognitive Prognosis Reasoning in Longitudinal 3D Brain MRI via Regional Volume Assessments
LoV3D: Vision-language model pipeline for longitudinal brain MRI analysis that grounds neurological disease progression reasoning in regional volume measurements.
3/17/2026
Revisiting Model Stitching In the Foundation Model Era
Research on model stitching technique for Vision Foundation Models, testing representational compatibility across models with different training objectives and data sources.
3/17/2026
From Text to Forecasts: Bridging Modality Gap with Temporal Evolution Semantic Space
Method for incorporating text into time-series forecasting by bridging the modality gap between qualitative text and quantitative forecasting signals through semantic space alignment.
3/17/2026
MetaKE: Meta-learning Aligned Knowledge Editing via Bi-level Optimization
MetaKE addresses knowledge editing in LLMs using bi-level optimization to fix specific facts without degrading general capabilities, identifying semantic-execution misalignment issues.
3/17/2026
Experimental evidence of progressive ChatGPT models self-convergence
Empirical study of model collapse in large language models trained recursively on synthetic data.
3/17/2026
Is Human Annotation Necessary? Iterative MBR Distillation for Error Span Detection in Machine Translation
Self-evolution framework using Minimum Bayes Risk decoding for error span detection in machine translation without human annotations.
3/17/2026
daVinci-Env: Open SWE Environment Synthesis at Scale
Large-scale open-source software engineering environment for training AI agents with executable, verifiable tasks and dynamic feedback.
3/17/2026
Continual Fine-Tuning with Provably Accurate and Parameter-Free Task Retrieval
Method for continual fine-tuning of pre-trained models on sequential tasks with parameter-free task retrieval and no forgetting.
3/17/2026
Your Code Agent Can Grow Alongside You with Structured Memory
Code agent framework with structured memory enabling adaptive learning from project evolution and past successful reasoning trajectories.
3/17/2026
Beyond Attention: True Adaptive World Models via Spherical Kernel Operator
World model architecture using spherical kernel operators to handle shifting data distributions in latent space transitions.
3/17/2026
Federated Personal Knowledge Graph Completion with Lightweight Large Language Models for Personalized Recommendations
Federated framework combining lightweight LLMs with personal knowledge graphs for privacy-preserving personalized recommendations.
3/17/2026
Knowledge, Rules and Their Embeddings: Two Paths towards Neuro-Symbolic JEPA
Neuro-symbolic architecture combining self-supervised learning with verifiable logic rules to mitigate spurious correlations and shortcut learning.
3/17/2026
Learning from Partial Chain-of-Thought via Truncated-Reasoning Self-Distillation
Self-distillation method reducing computational cost of chain-of-thought reasoning by training models to generate correct predictions from truncated reasoning.
3/17/2026
PREBA: Surgical Duration Prediction via PCA-Weighted Retrieval-Augmented LLMs and Bayesian Averaging Aggregation
Zero-shot LLM approach for surgical duration prediction combining retrieval-augmentation with Bayesian averaging, avoiding need for fine-tuning.
3/17/2026
FastODT: A tree-based framework for efficient continual learning
Tree-based continual learning framework for non-stationary data distributions with constrained computational resources in time series applications.
3/17/2026
Learning Retrieval Models with Sparse Autoencoders
Sparse autoencoders foundation for learned sparse retrieval, decomposing LLM representations into interpretable latent features for efficient document retrieval.
3/17/2026
ICaRus: Identical Cache Reuse for Efficient Multi Model Inference
Multi-model inference optimization reusing identical KV caches across models to reduce memory consumption in agentic AI systems.
3/17/2026
FedTreeLoRA: Reconciling Statistical and Functional Heterogeneity in Federated LoRA Fine-Tuning
Federated learning method for LoRA fine-tuning of LLMs addressing statistical and functional heterogeneity across model layers.
3/17/2026
Brittlebench: Quantifying LLM robustness via prompt sensitivity
Benchmark quantifying LLM robustness by measuring model sensitivity to prompt variations, typos, and paraphrases in real-world conditions.
3/17/2026
From Stochastic Answers to Verifiable Reasoning: Interpretable Decision-Making with LLM-Generated Code
Framework reframing LLMs as code generators for interpretable decision-making in high-stakes scenarios, improving reproducibility over black-box approaches.
3/17/2026
RelayCaching: Accelerating LLM Collaboration via Decoding KV Cache Reuse
KV cache optimization technique for multi-agent LLM systems that reuses decoding caches to reduce memory usage and latency in collaborative AI tasks.