3/3/2026
Spectral Condition for $\mu$P under Width-Depth Scaling
Spectral condition analysis for maximal update parameterization under joint width-depth scaling in foundation models.
3/3/2026
Learning to Attack: A Bandit Approach to Adversarial Context Poisoning
AdvBandit black-box adaptive attack on neural contextual bandits formulating context poisoning as continuous-armed bandit problem.
3/3/2026
DeepAFL: Deep Analytic Federated Learning
Analytic federated learning approach replacing gradient-based updates with closed-form solutions for improved convergence and scalability.
3/3/2026
Unlearning Evaluation through Subset Statistical Independence
Subset-level evaluation framework for machine unlearning using statistical independence without retraining or membership inference.
3/3/2026
Learning to Explore: Policy-Guided Outlier Synthesis for Graph Out-of-Distribution Detection
Policy-guided outlier synthesis approach for unsupervised out-of-distribution detection in graph neural networks.
3/3/2026
Multi-Domain Riemannian Graph Gluing for Building Graph Foundation Models
Multi-domain graph pre-training framework for building graph foundation models with theoretical analysis of knowledge transfer.
3/3/2026
IDER: IDempotent Experience Replay for Reliable Continual Learning
IDER method addressing catastrophic forgetting in continual learning with uncertainty calibration for mission-critical deployment.
3/3/2026
Adapt Data to Model: Adaptive Transformation Optimization for Domain-shared Time Series Foundation Models
Data-centric framework for adapting diverse time series to large time series models without retraining.
3/3/2026
Frozen Policy Iteration: Computationally Efficient RL under Linear $Q^{\pi}$ Realizability for Deterministic Dynamics
Frozen Policy Iteration algorithm for computationally efficient reinforcement learning under linear Q-function realizability.
3/3/2026
MARS: Harmonizing Multimodal Convergence via Adaptive Rank Search
MARS framework for efficient fine-tuning of multimodal LLMs using adaptive rank search to handle training imbalance across modalities.
3/3/2026
To Use or not to Use Muon: How Simplicity Bias in Optimizers Matters
Analysis of Muon optimizer's simplicity bias and potential downsides compared to Adam for neural network training.
3/3/2026
General Proximal Flow Networks
Generalizes Bayesian Flow Networks with arbitrary divergence/distance functions replacing fixed KL divergence, enabling broader belief-update operators.
3/3/2026
Identifying and Characterising Response in Clinical Trials: Development and Validation of a Machine Learning Approach in Colorectal Cancer
Combines partly conditional modeling with machine learning to identify patient response subgroups in colorectal cancer clinical trials using repeated measures.
3/3/2026
Interpretable Cross-Network Attention for Resting-State fMRI Representation Learning
Self-supervised learning framework for resting-state fMRI using masked reconstruction with cross-attention for interpretable brain network representation learning.
3/3/2026
Identifying the Geographic Foci of US Local News
Proposes geo-foci model for identifying salient geographical locations in US local news coverage to address economic pressures on local journalism.
3/3/2026
Neural Latent Arbitrary Lagrangian-Eulerian Grids for Fluid-Solid Interaction
Neural approach to fluid-solid interaction using latent arbitrary Lagrangian-Eulerian grids for capturing two-way nonlinear interactions in FSI problems.
3/3/2026
Lookahead identification in adversarial bandits: accuracy and memory bounds
Studies identification problem in adversarial multi-armed bandits for selecting arms performing best at future time with accuracy and memory bounds.
3/3/2026
Curation Leaks: Membership Inference Attacks against Data Curation for Machine Learning
Demonstrates membership inference attacks against data curation methods used for private ML, showing curation-based privacy solutions leak training data membership.
3/3/2026
A Gauge Theory of Superposition: Toward a Sheaf-Theoretic Atlas of Neural Representations
Develops gauge-theoretic framework for superposition in LLMs using sheaf-theoretic atlas replacing single-global-dictionary with local semantic charts and Fisher metrics.
3/3/2026
GeMi: A Graph-based, Multimodal Recommendation System for Narrative Scroll Paintings
Graph Neural Network-based recommendation system for multimodal data handling content-based recommendations with user preference incorporation.
3/3/2026
Navigating Time's Possibilities: Plausible Counterfactual Explanations for Multivariate Time-Series Forecast through Genetic Algorithms
Novel method for counterfactual learning in multivariate time series using genetic algorithms to uncover causal relationships and interventions.
3/3/2026
MultiPUFFIN: A Multimodal Domain-Constrained Foundation Model for Molecular Property Prediction of Small Molecules
Presents MultiPUFFIN, a domain-constrained multimodal foundation model for molecular property prediction ensuring thermodynamic consistency across chemical space.
3/3/2026
Active Flow Matching
Introduces Active Flow Matching combining discrete diffusion/flow models with variational frameworks for black-box optimization without retraining.
3/3/2026
Knowledge without Wisdom: Measuring Misalignment between LLMs and Intended Impact
Evaluates performance misalignment between leading LLMs on out-of-distribution educational tasks, finding inter-model behaviors correlate higher than with human expert behaviors.
3/3/2026
Probabilistic Learning and Generation in Deep Sequence Models
Addresses uncertainty awareness in deep sequence models by integrating Bayesian methods with probabilistic learning, comparing approximate inference techniques.
3/3/2026
Evaluating AI Grading on Real-World Handwritten College Mathematics: A Large-Scale Study Toward a Benchmark
Large-scale empirical study of AI grading on handwritten calculus work using OCR-conditioned LLMs with rubric-guided prompting for score and feedback generation.
3/3/2026
Principled Fast and Meta Knowledge Learners for Continual Reinforcement Learning
Proposes dual-learner framework combining fast learner and meta learner for continual RL, inspired by hippocampus-cortex interaction for knowledge transfer and integration.
3/3/2026
When Does Margin Clamping Affect Training Variance? Dataset-Dependent Effects in Contrastive Forward-Forward Learning
Investigates margin clamping effects on training variance in Contrastive Forward-Forward learning for Vision Transformers, addressing instability sources.
3/3/2026
Stabilizing Policy Optimization via Logits Convexity
Analyzes stability gap between supervised fine-tuning and RL in LLM training from gradient perspective, showing logits convexity role in stable optimization.
3/3/2026
Intent-Context Synergy Reinforcement Learning for Autonomous UAV Decision-Making in Air Combat
Proposes Intent-Context Synergy RL approach for autonomous UAV decision-making in contested environments, addressing trade-offs between mission efficiency and survivability.
3/3/2026
Forgetting is Competition: Rethinking Unlearning as Representation Interference in Diffusion Models
arXiv 2603.00975: Representation interference framework for selective unlearning in text-to-image diffusion models preserving quality.
3/3/2026
Compensation-free Machine Unlearning in Text-to-Image Diffusion Models by Eliminating the Mutual Information
arXiv 2603.00992: Machine unlearning method for diffusion models removing sensitive concepts via mutual information elimination.
3/3/2026
One-Token Verification for Reasoning Correctness Estimation
arXiv 2603.01025: One-token verification method for estimating correctness in LLM reasoning with reduced computational cost.
3/3/2026
Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift
arXiv 2603.01040: Fed-ADE for federated learning adaptation under distribution shifts without ground-truth labels.
3/3/2026
Evaluating GFlowNet from partial episodes for stable and flexible policy-based training
arXiv 2603.01047: GFlowNet training improvements via partial episodes for stable policy-based sampling of combinatorial candidates.
3/3/2026
No More Maybe-Arrows: Resolving Causal Uncertainty by Breaking Symmetries
arXiv 2603.01052: CausalSAGE framework for refining causal discovery PAGs into DAGs by breaking symmetries.
3/3/2026
Understanding LoRA as Knowledge Memory: An Empirical Analysis
arXiv 2603.01097: Empirical analysis of LoRA as parametric knowledge memory for continuous LLM updates without context constraints.
3/3/2026
Demystifying Group Relative Policy Optimization: Its Policy Gradient is a U-Statistic
arXiv 2603.01162: Theoretical analysis of GRPO through U-statistics lens, core method in DeepSeekMath and DeepSeek-R1 for LLM reasoning.
3/3/2026
SphUnc: Hyperspherical Uncertainty Decomposition and Causal Identification via Information Geometry
arXiv 2603.01168: SphUnc framework combining hyperspherical representation learning with causal modeling for uncertainty decomposition.
3/3/2026
PARWiS: Winner determination under shoestring budgets using active pairwise comparisons
arXiv 2603.01171: PARWiS algorithm for winner determination via active pairwise comparisons with reinforcement learning variant.
3/3/2026
Scaling of learning time for high dimensional inputs
arXiv 2603.01184: Theoretical analysis of learning time trade-offs for high-dimensional neural network inputs.
3/3/2026
Subliminal Signals in Preference Labels
arXiv 2603.01204: Research on LLM-as-judge frameworks showing preference labels can function as covert communication channels between models.
3/3/2026
Learn Hard Problems During RL with Reference Guided Fine-tuning
arXiv 2603.01223: RL method for LLM mathematical reasoning using reference solutions to overcome reward sparsity in hard problems.
3/3/2026
Show HN: AgentOx – MCP Security and Conformance Auditor
Open-source Rust CLI auditor for MCP servers, checking protocol conformance, security, and behavioral contracts before production deployment.
3/3/2026
AI Authentication and Authorization
Article on applying OAuth/API identity patterns to secure AI systems and agents with authentication/authorization.
3/3/2026
Ed Gutenburg: The First Autonomous Investigative Reporter
Proposal for autonomous investigative reporter agents that can conduct research, publish findings, and pressure institutions on behalf of individual users.
3/3/2026
Building an Open-Source Verilog Simulator with AI: 580K Lines in 43 Days
Engineer used AI agents to build open-source Verilog simulator with 580K lines in 43 days, including simulation, formal verification, and mutation testing capabilities.
3/3/2026
Detecting LLM-Generated Web Novels Using "Classical" Machine Learning (AIGC Text Detection)
ML technique for detecting LLM-generated text using classical machine learning models. Includes online demo.
3/3/2026
Do AI Agents Make Money in 2026? Or Is It Just Mac Minis and Vibes?
Investigation into AI agent monetization claims in 2026, examining reality behind Mac Mini setups and autonomous income stream claims versus hype.
3/3/2026
Shutting down, open sourced private AI document server
Open-source private document server using AI to answer questions about uploaded documents, with SQL database for structured data and local processing.