3/12/2026
Code-Space Response Oracles: Generating Interpretable Multi-Agent Policies with Large Language Models
Framework generating interpretable multi-agent policies using LLMs as game-theoretic response oracles for transparent RL policies.
3/12/2026
Adaptive Activation Cancellation for Hallucination Mitigation in Large Language Models
Inference-time framework treating hallucinations in LLMs as structured interference in transformer residual streams using adaptive activation cancellation.
3/12/2026
Why Does It Look There? Structured Explanations for Image Classification
Explainability method for image classification providing structured interpretations without relying on auxiliary models.
3/12/2026
One Adapter for All: Towards Unified Representation in Step-Imbalanced Class-Incremental Learning
Class-incremental learning method addressing step imbalance where task streams have varying numbers of classes.
3/12/2026
From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning
Distribution Contractive RL framework that finetunes pretrained diffusion/flow policies for robot control using online feedback.
3/12/2026
MultiwayPAM: Multiway Partitioning Around Medoids for LLM-as-a-Judge Score Analysis
MultiwayPAM clustering method for analyzing LLM-as-a-Judge scores to reduce computational cost and reveal evaluator bias structure.
3/12/2026
Quantum entanglement provides a competitive advantage in adversarial games
Study of quantum entanglement advantages in adversarial reinforcement learning games using quantum-classical hybrid agents.
3/12/2026
Hybrid Self-evolving Structured Memory for GUI Agents
GUI agents using vision-language models with hybrid self-evolving structured memory to handle long-horizon workflows and diverse interfaces.
3/12/2026
HEAL: Hindsight Entropy-Assisted Learning for Reasoning Distillation
HEAL distills reasoning from large language models to smaller models via hindsight entropy-assisted learning, overcoming teacher limitations.
3/12/2026
Adaptive Active Learning for Regression via Reinforcement Learning
WiGS formulates active learning sample selection as RL problem with dynamic additive criterion replacing static greedy sampling.
3/12/2026
VERI-DPO: Evidence-Aware Alignment for Clinical Summarization via Claim Verification and Direct Preference Optimization
VERI-DPO uses claim verification with DPO to improve clinical LLM summarization accuracy and reduce unsupported statements.
3/12/2026
Resource-constrained Amazons chess decision framework integrating large language models and graph attention
Hybrid framework combining LLMs and graph attention for Amazons chess in resource-constrained environments. Lightweight decision-making system.
3/12/2026
IH-Challenge: A Training Dataset to Improve Instruction Hierarchy on Frontier LLMs
IH-Challenge dataset for training instruction hierarchy in LLMs. Addresses jailbreaks and prompt injection by establishing trust-ordered instruction conflict resolution.
3/12/2026
Does LLM Alignment Really Need Diversity? An Empirical Study of Adapting RLVR Methods for Moral Reasoning
Empirical study of RLVR methods for LLM alignment in moral reasoning. Questions whether diversity is necessary for alignment tasks.
3/12/2026
Self-Scaled Broyden Family of Quasi-Newton Methods in JAX
JAX implementation of quasi-Newton optimization methods (BFGS, DFP, Broyden variants) built on Optimistix library.
3/12/2026
Detecting and Eliminating Neural Network Backdoors Through Active Paths with Application to Intrusion Detection
Novel method detecting and eliminating neural network backdoor triggers via active path analysis with intrusion detection application.
3/12/2026
FAME: Formal Abstract Minimal Explanation for Neural Networks
FAME: formal abductive explanations for neural networks using abstract interpretation with reduced explanation size for large models.
3/12/2026
EvoSchema: Towards Text-to-SQL Robustness Against Schema Evolution
EvoSchema addresses text-to-SQL robustness when database schemas evolve, improving model generalization across schema changes.
3/12/2026
CacheSolidarity: Preventing Prefix Caching Side Channels in Multi-tenant LLM Serving Systems
CacheSolidarity: defense mechanism preventing prefix caching side-channel attacks in multi-tenant LLM serving systems.
3/12/2026
ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learning
Deep generative models for synthetic tabular data synthesis using reinforcement learning to learn conditional distributions in low-data, imbalanced settings.
3/12/2026
Pointy - A Lightweight Transformer for Point Cloud Foundation Models
Pointy: Lightweight transformer-based foundation model for point cloud data trained on 39k point clouds, outperforming larger models with less supervision.
3/12/2026
ForwardFlow: Simulation only statistical inference using deep learning
ForwardFlow: Deep learning framework for simulation-only statistical inference using normalizing flows to approximate posterior distributions.
3/12/2026
Bayesian Optimization with Gaussian Processes to Accelerate Stationary Point Searches
Bayesian optimization with Gaussian processes to accelerate stationary point searches on potential energy surfaces using surrogate models.
3/12/2026
Efficient Bayesian Updates for Deep Active Learning via Laplace Approximations
Uses Laplace approximations for efficient Bayesian updates in deep active learning without retraining, addressing batch redundancy through diversity.
3/12/2026
An Updated Assessment of Reinforcement Learning for Macro Placement
Reassesses deep reinforcement learning for macro placement in chip design; develops stronger simulated annealing baselines and releases public benchmarks.
3/12/2026
Large Language Models for Travel Behavior Prediction
Applies large language models for travel behavior prediction through natural language reasoning frameworks as alternative to numerical models.
3/12/2026
Optimal Transport Aggregation for Distributed Mixture-of-Experts
Optimal transport-based aggregation method for combining locally trained mixture-of-experts models from decentralized datasets.
3/12/2026
Communication-Efficient Multimodal Federated Learning: Joint Modality and Client Selection
Addresses communication efficiency in multimodal federated learning with joint modality and client selection for heterogeneous network settings.
3/12/2026
Class Incremental Learning with Task-Specific Batch Normalization and Out-of-Distribution Detection
Addresses catastrophic forgetting in incremental learning for image classification using task-specific batch normalization and out-of-distribution detection.
3/12/2026
Boosting Cross-problem Generalization in Diffusion-Based Neural Combinatorial Solver via Inference Time Adaptation
Improves generalization of diffusion-based neural combinatorial optimization solvers across problem scales using inference-time adaptation for NP-complete problems.
3/12/2026
Is CLIP ideal? No. Can we fix it? Yes!
Addresses fundamental geometric limitations of CLIP's multimodal latent space for handling complex visual-textual interactions through geometric improvements.
3/12/2026
An Algorithm to perform Covariance-Adjusted Support Vector Classification in Non-Euclidean Spaces
Extends Support Vector Machine classification to non-Euclidean spaces by analyzing limitations of max-margin classification and KKT boundary conditions in non-Euclidean geometry.
3/12/2026
Panda: A pretrained forecast model for chaotic dynamics
Panda: Pretrained foundation model for forecasting chaotic dynamical systems like fluid flows and neuronal activity using dynamical systems theory.
3/12/2026
CARTGen-IR: Synthetic Tabular Data Generation for Imbalanced Regression
CARTGen-IR: Deep generative model approach for synthetic tabular data generation to address imbalanced regression problems without arbitrary classification thresholds.
3/12/2026
Comparative Analysis of Modern Machine Learning Models for Retail Sales Forecasting
Comparative study evaluating statistical, tree-based, and deep learning models (XGBoost, LightGBM, N-BEATS, Temporal Fusion Transformer) for retail sales forecasting with intermittent demand.
3/12/2026
Sequential-Parallel Duality in Prefix Scannable Models
Characterizes neural sequence models supporting sequential-parallel duality, analyzing architectures like Gated Linear Attention and Mamba.
3/12/2026
Global Minimizers of Sigmoid Contrastive Loss
Theoretical analysis of sigmoid contrastive loss in CLIP-style models, explaining advantages of temperature and bias in SigLIP and SigLIP2.
3/12/2026
One-Prompt Strikes Back: Sparse Mixture of Experts for Prompt-based Continual Learning
Sparse mixture-of-experts approach for prompt-based continual learning reducing computational overhead and memory scaling with number of tasks.
3/12/2026
Composer: A Search Framework for Hybrid Neural Architecture Design
Search framework for automatically discovering hybrid neural architectures combining attention, MLP, and other primitives beyond standard transformers.
3/12/2026
Communication Enables Cooperation in LLM Agents: A Comparison with Curriculum-Based Approaches
Compares communication versus curriculum learning for multi-agent LLM cooperation in game-theoretic settings, showing cheap talk achieves 96.7% cooperation.
3/12/2026
Reveal-to-Revise: Explainable Bias-Aware Generative Modeling with Multimodal Attention
Explainable generative framework combining cross-modal attention, Grad-CAM++ attribution, and bias-aware training for image generation auditing.
3/12/2026
Absolute indices for determining compactness, separability and number of clusters
Proposes absolute cluster validity indices to determine compactness, separability and optimal cluster count independent of algorithm comparison.
3/12/2026
Predicting kernel regression learning curves from only raw data statistics
Theoretical framework predicting kernel regression learning curves from data covariance and target function decomposition on real datasets.
3/12/2026
Revisiting Value Iteration: Unified Analysis of Discounted and Average-Reward Cases
Unified convergence analysis of value iteration algorithm for both discounted and average-reward reinforcement learning settings.
3/12/2026
STREAM-VAE: Dual-Path Routing for Slow and Fast Dynamics in Vehicle Telemetry Anomaly Detection
Variational autoencoder with dual-path routing for detecting anomalies in vehicle telemetry with mixed slow drifts and fast spikes.
3/12/2026
Hierarchical Dual-Strategy Unlearning for Biomedical and Healthcare Intelligence Using Imperfect and Privacy-Sensitive Medical Data
Hierarchical dual-strategy framework for selective unlearning in LLMs to remove privacy-sensitive healthcare information while preserving medical competencies.
3/12/2026
Partially Equivariant Reinforcement Learning in Symmetry-Breaking Environments
Applies group-equivariant inductive bias to reinforcement learning for environments with partial symmetry breaking.
3/12/2026
Pretrained battery transformer (PBT): A foundation model for universal battery life prediction
Foundation model using transformers for battery cycle life prediction across heterogeneous chemistries and operating conditions via transfer learning.
3/12/2026
Saddle-to-Saddle Dynamics Explains A Simplicity Bias Across Neural Network Architectures
Theoretical framework explaining simplicity bias in neural networks through saddle-to-saddle learning dynamics across architectures.
3/12/2026
Geometric Scaling of Bayesian Inference in LLMs
Investigates geometric substrate of Bayesian inference in production-scale language models (Pythia, Phi-2, Llama-3, Mistral).