3/23/2026
TTQ: Activation-Aware Test-Time Quantization to Accelerate LLM Inference On The Fly
TTQ: test-time quantization framework for LLM inference acceleration without retraining, addressing domain shift in calibration.
3/23/2026
CLaRE-ty Amid Chaos: Quantifying Representational Entanglement to Predict Ripple Effects in LLM Editing
CLaRE quantifies representational entanglement to predict unintended side effects from LLM editing techniques.
3/23/2026
Parameter-Efficient Token Embedding Editing for Clinical Class-Level Unlearning
STEU enables parameter-efficient machine unlearning for clinical language models through sparse token embedding editing.
3/23/2026
MemReward: Graph-Based Experience Memory for LLM Reward Prediction with Limited Labels
MemReward uses graph-based experience memory for LLM reward prediction with limited labels, addressing expensive human annotation costs.
3/23/2026
LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels
LeWorldModel: stable end-to-end Joint Embedding Predictive Architecture for learning world models from raw pixels without auxiliary supervision.
3/23/2026
MSNet and LS-Net: Scalable Multi-Scale Multi-Representation Networks for Time Series Classification
MSNet and LS-Net: scalable multi-scale convolutional architectures for time series classification with multi-representation inputs.
3/23/2026
Ternary Gamma Semirings: From Neural Implementation to Categorical Foundations
Theoretical framework connecting neural network learning with Ternary Gamma Semirings to improve compositional generalization.
3/23/2026
Do Post-Training Algorithms Actually Differ? A Controlled Study Across Model Scales Uncovers Scale-Dependent Ranking Inversions
OXRL framework evaluates 51 post-training alignment algorithms across model scales, revealing scale-dependent ranking differences among DPO, SimPO, KTO, GRPO.
3/23/2026
DAPA: Distribution Aware Piecewise Activation Functions for On-Device Transformer Inference and Training
Proposes DAPA, a hardware-friendly activation function for on-device Transformer inference and training optimized for energy efficiency.
3/23/2026
Beyond Weighted Summation: Learnable Nonlinear Aggregation Functions for Robust Artificial Neurons
Research on replacing weighted summation with learnable nonlinear aggregation functions in neural networks to improve robustness against noisy inputs.
3/23/2026
Anatomical Heterogeneity in Transformer Language Models
Empirical analysis of transformer layer heterogeneity in SmolLM2-135M using diagnostic metrics including weight predictability and ablation degradation.
3/23/2026
A Mathematical Theory of Understanding
Mathematical framework for information absorption and understanding, relating signal clarity to learner structural capacity.
3/23/2026
Warm-Start Flow Matching for Guaranteed Fast Text/Image Generation
Flow matching approach with warm-start initialization for faster inference in autoregressive LLMs and diffusion models.
3/23/2026
Optimizing Resource-Constrained Non-Pharmaceutical Interventions for Multi-Cluster Outbreak Control Using Hierarchical Reinforcement Learning
Hierarchical reinforcement learning for allocating scarce public health resources across multiple disease outbreak clusters.
3/23/2026
GeoLAN: Geometric Learning of Latent Explanatory Directions in Large Language Models
Framework for interpreting LLM token representations using geometric learning and regularizers inspired by harmonic analysis.
3/23/2026
Deep Hilbert--Galerkin Methods for Infinite-Dimensional PDEs and Optimal Control
Deep learning approximation methods for nonlinear PDEs on Hilbert spaces using neural operators with universal approximation guarantees.
3/23/2026
Global Convergence of Multiplicative Updates for the Matrix Mechanism: A Collaborative Proof with Gemini 3
Convergence analysis of multiplicative updates in regularized nuclear norm optimization for private machine learning.
3/23/2026
Adaptive Layerwise Perturbation: Unifying Off-Policy Corrections for LLM RL
Off-policy correction technique for LLM reinforcement learning addressing policy staleness and training-inference distribution mismatch.
3/23/2026
TRACE: Trajectory Recovery with State Propagation Diffusion for Urban Mobility
Diffusion-based method for recovering dense trajectories from sparse GPS data in urban mobility applications.
3/23/2026
Any-Subgroup Equivariant Networks via Symmetry Breaking
Neural network architectures with flexible equivariance to arbitrary symmetries for diverse data modalities.
3/23/2026
ICLAD: In-Context Learning for Unified Tabular Anomaly Detection Across Supervision Regimes
In-context learning approach for tabular anomaly detection across multiple supervision regimes without retraining.
3/23/2026
Stochastic Sequential Decision Making over Expanding Networks with Graph Filtering
Graph filtering for decision-making on expanding networks with unknown attachment patterns and evolving topology.
3/23/2026
Scalable Cross-Facility Federated Learning for Scientific Foundation Models on Multiple Supercomputers
Federated learning approach for training large foundation models on distributed supercomputer facilities with privacy and data sovereignty constraints.
3/23/2026
Subspace Kernel Learning on Tensor Sequences
Kernel learning framework for tensor data with mode-wise subspace comparison for large-scale structured multi-way data.
3/23/2026
Neural Uncertainty Principle: A Unified View of Adversarial Fragility and LLM Hallucination
Unified geometric framework explaining adversarial vulnerability in vision models and hallucinations in LLMs via uncertainty principle.
3/23/2026
Wearable Foundation Models Should Go Beyond Static Encoders
Foundation models for wearable devices moving beyond static encoders toward temporal reasoning for health monitoring.
3/23/2026
ARMOR: Adaptive Resilience Against Model Poisoning Attacks in Continual Federated Learning for Mobile Indoor Localization
Defense mechanism against model poisoning attacks in continual federated learning for indoor localization systems.
3/23/2026
Demonstrations, CoT, and Prompting: A Theoretical Analysis of ICL
Theoretical analysis of in-context learning in LLMs, examining demonstration selection, chain-of-thought prompting, and adaptation without parameter updates.
3/23/2026
On Performance Guarantees for Federated Learning with Personalized Constraints
Federated learning with personalized constraints for heterogeneous agents in distributed optimization problems.
3/23/2026
DeepStock: Reinforcement Learning with Policy Regularizations for Inventory Management
Deep reinforcement learning with policy regularizations for inventory management, improving training stability and hyperparameter sensitivity.
3/23/2026
Continual Learning for Food Category Classification Dataset: Enhancing Model Adaptability and Performance
Continual learning framework for food image classification enabling incremental category addition without retraining from scratch.
3/23/2026
Alternating Diffusion for Proximal Sampling with Zeroth Order Queries
Proximal sampling method using zeroth-order function queries and simulated diffusion dynamics, avoiding score estimation requirements.
3/23/2026
Heavy-Tailed and Long-Range Dependent Noise in Stochastic Approximation: A Finite-Time Analysis
Finite-time convergence analysis of stochastic approximation under heavy-tailed and long-range dependent noise for optimization and reinforcement learning.
3/23/2026
Ensembles-based Feature Guided Analysis
Ensemble-based extension of Feature Guided Analysis for improving recall of DNN behavior explanations beyond precision limitations of existing FGA methods.
3/23/2026
The Residual Stream Is All You Need: On the Redundancy of the KV Cache in Transformer Inference
Proves KV cache in transformer inference is mathematically redundant; keys and values are deterministic residual stream projections with zero reconstruction error.
3/23/2026
Scale-Dependent Radial Geometry and Metric Mismatch in Wasserstein Propagation for Reverse Diffusion
Theoretical analysis of reverse diffusion sampling under Wasserstein geometry, identifying scale-dependent radial contraction properties.
3/23/2026
GoAgent: Group-of-Agents Communication Topology Generation for LLM-based Multi-Agent Systems
GoAgent generates task-specific communication topologies for LLM-based multi-agent systems to optimize coordination and group-based problem-solving.
3/23/2026
Regret Analysis of Sleeping Competing Bandits
Theoretical analysis of multi-armed bandits with competing players and time-varying availability, extending regret bounds to sleeping competing bandits framework.
3/23/2026
Learning from Similarity/Dissimilarity and Pairwise Comparison
Binary classification method using weak pairwise labels (similarity/dissimilarity) instead of explicit instance-level labels through probabilistic supervision framework.
3/23/2026
FedRG: Unleashing the Representation Geometry for Federated Learning with Noisy Clients
FedRG addresses noisy label robustness in federated learning by leveraging representation geometry rather than scalar loss values for sample identification.
3/23/2026
FedPDPO: Federated Personalized Direct Preference Optimization for Large Language Model Alignment
FedPDPO framework applies Direct Preference Optimization to federated learning for LLM alignment with decentralized, non-IID preference data while preserving privacy.
3/23/2026
Dual Path Attribution: Efficient Attribution for SwiGLU-Transformers through Layer-Wise Target Propagation
Novel attribution method (DPA) for understanding transformer LLM internals through efficient layer-wise target propagation, addressing computational costs of dense component attribution.
3/23/2026
Scalable Learning of Multivariate Distributions via Coresets
Coreset construction method for scalable multivariate conditional transformation models and density estimation.
3/23/2026
Two-Time-Scale Learning Dynamics: A Population View of Neural Network Training
Theoretical framework analyzing two-time-scale learning dynamics in population-based neural network training methods.
3/23/2026
FIPO: Eliciting Deep Reasoning with Future-KL Influenced Policy Optimization
FIPO is an RL algorithm improving token-level credit assignment in LLM reasoning tasks beyond outcome-based reward methods.
3/23/2026
NASimJax: GPU-Accelerated Policy Learning Framework for Penetration Testing
NASimJax is a GPU-accelerated framework for training RL policies on penetration testing with large action spaces.
3/23/2026
What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time
Test-time RL method for LLMs using selective-complementary reinforcement learning to improve reasoning under weak consensus conditions.
3/23/2026
Integrating Meta-Features with Knowledge Graph Embeddings for Meta-Learning
Meta-learning approach integrating meta-features with knowledge graph embeddings for pipeline performance prediction and dataset similarity.
3/23/2026
Memori: A Persistent Memory Layer for Efficient, Context-Aware LLM Agents
Memori is an LLM-agnostic persistent memory layer for context-aware AI agents supporting multi-session interactions without vendor lock-in.
3/23/2026
Model-Driven Learning-Based Physical Layer Authentication for Mobile Wi-Fi Devices
Model-driven learning for physical layer authentication in wireless IoT devices using hybrid approach.