3/23/2026
torchgfn: A PyTorch GFlowNet library
Proposes pixel-level benchmark and taxonomy for VLM image tampering detection. Reformulates task with language-aware metrics beyond coarse region labels.
3/23/2026
Causal Learning in Biomedical Applications: Krebs Cycle as a Benchmark
torchgfn is a PyTorch library for generative flow networks with modular architecture. Facilitates testing new GFlowNet methods and training approaches.
3/23/2026
Simulation-based Inference with the Python Package sbijax
Python package sbijax implementing neural simulation-based inference methods for Bayesian inference with intractable likelihoods.
3/23/2026
Online Clustering of Data Sequences with Bandit Information
Online clustering algorithm for data sequences using multi-armed bandit framework with fixed-confidence setting.
3/23/2026
Flow-based Conformal Prediction for Multi-dimensional Time Series
Conformal prediction method for multi-dimensional time series uncertainty quantification using flow-based models.
3/23/2026
CatBOX: A Categorical-Continuous Bayesian Optimization with Spectral Mixture Kernels for Accelerated Catalysis Experiments
Bayesian optimization method with spectral mixture kernels for accelerated catalyst discovery combining categorical and continuous parameters.
3/23/2026
HALO: Hierarchical Reinforcement Learning for Large-Scale Adaptive Traffic Signal Control
Hierarchical reinforcement learning approach for scalable adaptive traffic signal control in city-scale smart city environments.
3/23/2026
Hidden Breakthroughs in Language Model Training
Analysis method to identify hidden breakthroughs in language model training by analyzing loss curves beyond single scalar metrics.
3/23/2026
Predictive Scaling Laws for Efficient GRPO Training of Large Reasoning Models
Predictive framework deriving scaling laws for efficient GRPO training of reasoning LLMs to optimize resource usage during fine-tuning.
3/23/2026
Fast 3D Diffusion for Scalable Granular Media Synthesis
Generative pipeline using 3D diffusion models to synthesize large-scale granular media assemblies in mechanically realistic configurations.
3/23/2026
FNODE: Flow-Matching for data-driven simulation of constrained multibody systems
Flow-Matching Neural ODE framework for data-driven simulation of constrained multibody dynamics with supervised acceleration training.
3/23/2026
An upper bound on the silhouette evaluation metric for clustering
Theoretical analysis deriving upper bounds on silhouette coefficient for clustering quality evaluation.
3/23/2026
Does Weak-to-strong Generalization Happen under Spurious Correlations?
Theoretical and algorithmic study of weak-to-strong generalization with pseudolabels under spurious correlations from group imbalance.
3/23/2026
Less is More: Towards Simple Graph Contrastive Learning
Simplified graph contrastive learning method for unsupervised representation learning on heterophilic graphs without complex augmentation schemes.
3/23/2026
Gym-TORAX: Open-source software for integrating reinforcement learning with plasma control simulators in tokamak research
Open-source Python package integrating reinforcement learning with tokamak plasma control simulators using Gymnasium environments.
3/23/2026
KoALA: KL-L0 Adversarial Detector via Label Agreement
Adversarial detection method for neural networks requiring no architectural changes, detecting attacks through complementary classifier label agreement.
3/23/2026
ACT as Human: Multimodal Large Language Model Data Annotation with Critical Thinking
Data annotation pipeline using LLMs with critical thinking as both annotators and judges to improve supervised learning label quality.
3/23/2026
Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive Drafter
Efficient RL training method for reasoning LLMs using adaptive drafting to handle long-tail response generation distribution and reduce computation time.
3/23/2026
Efficient Cross-Domain Offline Reinforcement Learning with Dynamics- and Value-Aligned Data Filtering
Cross-domain offline reinforcement learning method using dynamics and value alignment for filtering datasets to improve agent training in target environments.
3/23/2026
ReLaX: Reasoning with Latent Exploration for Large Reasoning Models
ReLaX: approach addressing entropy collapse in large reasoning models by promoting latent-level exploration during reinforcement learning with verifiable rewards.
3/23/2026
Bridging Training and Merging Through Momentum-Aware Optimization
Unified framework maintaining factorized momentum states across neural network training and model merging to reduce redundant computation.
3/23/2026
Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models
Self-Distilled Reasoner: on-policy self-distillation approach for LLM reasoning that addresses distribution mismatch without teacher models.
3/23/2026
Reinforcement Unlearning via Group Relative Policy Optimization
Reinforcement Unlearning via GRPO: technique for removing sensitive data from LLMs without retraining, compliant with GDPR and EU AI Act.
3/23/2026
A Sheaf-Theoretic and Topological Perspective on Complex Network Modeling and Attention Mechanisms in Graph Neural Models
Sheaf-theoretic and topological perspective on signal diffusion and attention mechanisms in graph neural networks and geometric deep learning.
3/23/2026
How Understanding Forecast Uncertainty Resolves the Explainability Problem in Machine Learning Models
Analysis of forecast uncertainty in machine learning explainability, addressing instability of LIME and SHAP near decision boundaries.
3/23/2026
StealthRL: Reinforcement Learning Paraphrase Attacks for Multi-Detector Evasion of AI-Text Detectors
StealthRL: RL framework using group relative policy optimization to test robustness of AI-text detectors against adversarial paraphrasing attacks.
3/23/2026
A Task-Centric Theory for Iterative Self-Improvement with Easy-to-Hard Curricula
Theoretical analysis of iterative self-improvement in LLMs using reward-verified outputs with easy-to-hard curriculum learning.
3/23/2026
A Pragmatic Method for Comparing Clusterings with Overlaps and Outliers
Method for comparing clustering algorithms with overlapping clusters and outliers in unsupervised learning evaluation.
3/23/2026
Spectral Convolution on Orbifolds for Geometric Deep Learning
Spectral convolution techniques for geometric deep learning on non-Euclidean data structures like graphs and manifolds.
3/23/2026
Federated Learning Playground
Interactive browser-based platform teaching federated learning concepts with real-time visualization of heterogeneous data and aggregation algorithms.
3/23/2026
mlx-vis: GPU-Accelerated Dimensionality Reduction and Visualization on Apple Silicon
mlx-vis: GPU-accelerated dimensionality reduction library for Apple Silicon implementing 8 methods with hardware-accelerated rendering.
3/23/2026
FEAT: A Linear-Complexity Foundation Model for Extremely Large Structured Data
FEAT: linear-complexity foundation model for structured data handling heterogeneous datasets with improved attention mechanisms for large-scale applications.
3/23/2026
On the Cone Effect and Modality Gap in Medical Vision-Language Embeddings
Study of cone effect and modality gap in medical vision-language models, analyzing embedding concentration and cross-modal separation in supervised learning.
3/23/2026
AcceRL: A Distributed Asynchronous Reinforcement Learning and World Model Framework for Vision-Language-Action Models
AcceRL: distributed asynchronous RL framework for Vision-Language-Action models with integrated trainable world models, eliminating synchronization barriers.
3/23/2026
Balancing the Reasoning Load: Difficulty-Differentiated Policy Optimization with Length Redistribution for Efficient and Robust Reinforcement Learning
Difficulty-Differentiated Policy Optimization addresses Large Reasoning Models' overthinking and overconfidence by redistributing token allocation based on problem difficulty.
3/23/2026
LISAA: A Framework for Large Language Model Information Security Awareness Assessment
Framework for assessing information security awareness in LLMs, including security knowledge, attitudes, and behavior to improve rejection of unsafe requests.
3/23/2026
A Phylogenetic Approach to Genomic Language Modeling
Genomic language model framework using phylogenetic trees and multispecies alignment for identifying evolutionarily constrained sequences.
3/23/2026
Understanding and Optimizing Multi-Stage AI Inference Pipelines
Analysis of multi-stage LLM inference pipelines including RAG, KV cache retrieval, routing, and reasoning with optimization strategies.
3/23/2026
Pseudo-Simulation for Autonomous Driving
Pseudo-simulation method for evaluating autonomous vehicles addressing limitations of real-world and closed-loop simulation evaluation.
3/23/2026
Principled Multimodal Representation Learning
Framework for multimodal representation learning through simultaneous alignment of diverse data modalities.
3/23/2026
Superclass-Guided Representation Disentanglement for Spurious Correlation Mitigation
Method leveraging superclasses for representation disentanglement to mitigate spurious correlations and improve group robustness.
3/23/2026
Evaluation-Aware Reinforcement Learning
Evaluation-Aware RL framework considers policy evaluation accuracy during training to reduce variance and bias.
3/23/2026
Discovering Intersectional Bias via Directional Alignment in Face Recognition Embeddings
Method for detecting intersectional bias in face recognition embeddings using directional alignment in latent space.
3/23/2026
CARES: Context-Aware Resolution Selector for VLMs
CARES lightweight module selects appropriate image resolution for vision-language models to reduce token overhead and latency.
3/23/2026
RobotArena $\infty$: Scalable Robot Benchmarking via Real-to-Sim Translation
RobotArena∞ enables scalable robot benchmarking through real-to-sim translation for evaluating diverse robotic agents.
3/23/2026
Rep2Text: Decoding Full Text from a Single LLM Token Representation
Rep2Text framework recovers original input text from single LLM token representation using trainable adapter for interpretability.
3/23/2026
FastMMoE: Accelerating Multimodal Large Language Models through Dynamic Expert Activation and Routing-Aware Token Pruning
FastMMoE accelerates multimodal LLM inference through dynamic expert activation and token pruning for reduced latency.
3/23/2026
Unsupervised Feature Selection via Robust Autoencoder and Adaptive Graph Learning
Unsupervised feature selection method using robust autoencoder and adaptive graph learning for high-dimensional data clustering.
3/23/2026
Dementia-R1: Reinforced Pretraining and Reasoning from Unstructured Clinical Notes for Real-World Dementia Prognosis
Dementia-R1 applies reinforced pretraining and reasoning to LLMs for longitudinal clinical prognosis from unstructured medical notes.
3/23/2026
A Multi-Perspective Benchmark and Moderation Model for Evaluating Safety and Adversarial Robustness
Benchmark and moderation model for evaluating LLM safety, adversarial robustness, and handling of nuanced harmful content detection.