3/24/2026
Reward Sharpness-Aware Fine-Tuning for Diffusion Models
Sharpness-aware fine-tuning approach for diffusion models to reduce reward hacking in reinforcement learning from human feedback.
3/24/2026
Prompt replay: speeding up grpo with on-policy reuse of high-signal prompts
Online data selection method for GRPO reinforcement learning that reuses high-signal prompts to improve LLM reasoning training efficiency.
3/24/2026
ALMAB-DC: Active Learning, Multi-Armed Bandits, and Distributed Computing for Sequential Experimental Design and Black-Box Optimization
Bayesian sequential design framework combining active learning, multi-armed bandits, and distributed computing for black-box optimization.
3/24/2026
On the Role of Batch Size in Stochastic Conditional Gradient Methods
Theoretical analysis of batch size effects in stochastic conditional gradient optimization methods.
3/24/2026
Pretrained Video Models as Differentiable Physics Simulators for Urban Wind Flows
Pretrained video diffusion model repurposed as differentiable physics simulator for urban wind flow prediction.
3/24/2026
Does Mechanistic Interpretability Transfer Across Data Modalities? A Cross-Domain Causal Circuit Analysis of Variational Autoencoders
Analysis of mechanistic interpretability in VAEs across image and tabular data modalities using causal circuit analysis.
3/24/2026
Amortized Variational Inference for Logistic Regression with Missing Covariates
Amortized variational inference method for logistic regression with missing covariate data using VAE-based approach.
3/24/2026
Aggregation Alignment for Federated Learning with Mixture-of-Experts under Data Heterogeneity
Federated learning framework for fine-tuning Mixture-of-Experts LLMs on distributed data with privacy preservation.
3/24/2026
Fusing Memory and Attention: A study on LSTM, Transformer and Hybrid Architectures for Symbolic Music Generation
Comparative study of LSTM, Transformer, and hybrid architectures for symbolic music generation tasks.
3/24/2026
Sonny: Breaking the Compute Wall in Medium-Range Weather Forecasting
Data-driven weather forecasting using deep learning with reduced computational requirements compared to existing models.
3/24/2026
Direct Interval Propagation Methods using Neural-Network Surrogates for Uncertainty Quantification in Physical Systems Surrogate Model
Neural network surrogates for uncertainty quantification in physical systems through interval propagation methods.
3/24/2026
FluidWorld: Reaction-Diffusion Dynamics as a Predictive Substrate for World Models
Research on world models using reaction-diffusion dynamics as alternative to Transformers for predicting future environment states with better spatial inductive bias.
3/24/2026
Stream separation improves Bregman conditioning in transformers
Analysis of Bregman geometry in transformer representations, showing how stream separation improves steering methods.
3/24/2026
Active Inference Agency Formalization, Metrics, and Convergence Assessments
Formal framework for defining and analyzing agency in AI systems through continuous representation and mesa-optimization dynamics.
3/24/2026
AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Search
AutoKernel: open-source autonomous agent framework for GPU kernel optimization using iterative search on PyTorch models.
3/24/2026
The Workload-Router-Pool Architecture for LLM Inference Optimization: A Vision Paper from the vLLM Semantic Router Project
vLLM Semantic Router architecture for LLM inference optimization covering routing, caching, safety, and adaptive mechanisms.
3/24/2026
TIDE: Token-Informed Depth Execution for Per-Token Early Exit in LLM Inference
TIDE: post-training system with learned routers for per-token early exit in LLM inference, no retraining required.
3/24/2026
PLR: Plackett-Luce for Reordering In-Context Learning Examples
PLR: method using Plackett-Luce ranking to efficiently reorder in-context learning examples without exhaustive search.
3/24/2026
Constrained Online Convex Optimization with Memory and Predictions
Algorithms for constrained online convex optimization with memory constraints and predictions.
3/24/2026
A Generalised Exponentiated Gradient Approach to Enhance Fairness in Binary and Multi-class Classification Tasks
Fairness improvement method using exponentiated gradient approach for multi-class classification tasks to mitigate bias.
3/24/2026
Mechanisms of Introspective Awareness
Study of introspective awareness mechanisms in LLMs, investigating whether steering detection reflects genuine circuitry or shallow heuristics.
3/24/2026
DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment
DSPA: inference-time method using sparse autoencoders for LLM preference alignment without weight updates, enabling mechanistic steering.
3/24/2026
Off-Policy Evaluation for Ranking Policies under Deterministic Logging Policies
Off-policy evaluation methods for ranking systems using offline logged data, addressing bias in inverse propensity score estimators.
3/24/2026
Learning Can Converge Stably to the Wrong Belief under Latent Reliability
Research on how learning systems can converge to incorrect solutions when feedback reliability is unobservable, addressing theoretical issues in optimization.
3/24/2026
Multinoulli Extension: A Lossless Continuous Relaxation for Partition-Constrained Subset Selection
Continuous relaxation method for partition-constrained subset selection with submodular objectives, improving query complexity over existing local-search approaches.
3/24/2026
Quotient Geometry, Effective Curvature, and Implicit Bias in Simple Shallow Neural Networks
Develops differential-geometric framework accounting for parameter redundancy in shallow neural networks via quotient geometry to measure intrinsic predictor properties.
3/24/2026
Optimizing Feature Extraction for On-device Model Inference with User Behavior Sequences
Addresses on-device ML inference bottleneck by optimizing feature extraction from user behavior sequences for low-latency mobile app execution.
3/24/2026
BOxCrete: A Bayesian Optimization Open-Source AI Model for Concrete Strength Forecasting and Mix Optimization
Open-source Bayesian optimization model for concrete strength prediction and mix design optimization, applying ML to materials science with public datasets.
3/24/2026
Sharper Generalization Bounds for Transformer
Derives sharper generalization error bounds for Transformer architectures using offset Rademacher complexity across single and multi-head, multi-layer variants.
3/24/2026
What Do World Models Learn in RL? Probing Latent Representations in Learned Environment Simulators
Interpretability study probing internal representations of world models (IRIS and DIAMOND) in RL using linear/nonlinear probing and causal interventions.
3/24/2026
Kolmogorov Complexity Bounds for LLM Steganography and a Perplexity-Based Detection Proxy
Information-theoretic analysis of LLM steganography showing Kolmogorov complexity bounds on hidden payload embedding in text while preserving semantic meaning.
3/24/2026
SSAM: Singular Subspace Alignment for Merging Multimodal Large Language Models
SSAM method merges multiple pre-trained multimodal LLMs without additional training by aligning singular subspaces, enabling efficient multi-modality integration.
3/24/2026
In-network Attack Detection with Federated Deep Learning in IoT Networks: Real Implementation and Analysis
Lightweight autoencoder-based anomaly detection using federated learning for IoT networks, enabling privacy-preserving security monitoring on resource-constrained devices.
3/24/2026
Riemannian Geometry Speaks Louder Than Words: From Graph Foundation Model to Next-Generation Graph Intelligence
Framework for building general-purpose Graph Foundation Models using Riemannian geometry principles, analogous to large language models for graph-structured data.
3/24/2026
mSFT: Addressing Dataset Mixtures Overfiting Heterogeneously in Multi-task SFT
mSFT algorithm for multi-task supervised fine-tuning that addresses heterogeneous overfitting by dynamically adjusting compute budget per dataset to balance learning rates.
3/24/2026
Rule-State Inference (RSI): A Bayesian Framework for Compliance Monitoring in Rule-Governed Domains
Bayesian framework for compliance monitoring in rule-governed domains, inferring latent states given known rules rather than learning rules from data.
3/24/2026
Towards Multimodal Time Series Anomaly Detection with Semantic Alignment and Condensed Interaction
Multimodal time series anomaly detection model combining numerical and semantic data with alignment and interaction mechanisms for dynamic system monitoring.
3/24/2026
Proximal Policy Optimization in Path Space: A Schr\"odinger Bridge Perspective
GSB-PPO extends proximal policy optimization to trajectory-level generative policies using Schrödinger Bridge perspective, enabling diffusion and flow-based policy optimization.
3/24/2026
MISApp: Multi-Hop Intent-Aware Session Graph Learning for Next App Prediction
Session-based graph learning model for predicting next mobile app launches by modeling multi-hop intent patterns and handling sparse/cold-start user profiles.
3/24/2026
TrustFed: Enabling Trustworthy Medical AI under Data Privacy Constraints
Federated learning framework for privacy-preserving medical AI training across healthcare institutions while addressing data heterogeneity and deployment challenges.
3/24/2026
Data-Free Layer-Adaptive Merging via Fisher Information for Long-to-Short Reasoning LLMs
Model merging technique using Fisher Information to combine long-chain-of-thought and base LLMs, preserving reasoning accuracy while reducing output length without additional training.
3/24/2026
When Exploration Comes for Free with Mixture-Greedy: Do we need UCB in Diversity-Aware Multi-Armed Bandits?
Multi-armed bandit approach for selecting among generative models under diversity-aware metrics, addressing efficient model selection in generative AI without relying on classical UCB algorithms.
3/24/2026
Uncertainty Quantification for Distribution-to-Distribution Flow Matching in Scientific Imaging
arXiv paper on uncertainty quantification for distribution-to-distribution flow matching in scientific imaging applications.
3/24/2026
FISformer: Replacing Self-Attention with a Fuzzy Inference System in Transformer Models for Time Series Forecasting
FISformer replaces self-attention with fuzzy inference systems in transformers for time series forecasting, addressing uncertainty modeling limitations of dot-product attention.
3/24/2026
CellFluxRL: Biologically-Constrained Virtual Cell Modeling via Reinforcement Learning
Post-training virtual cell models with RL using biologically-constrained reward functions for drug discovery simulation.
3/24/2026
Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors
Precipitation nowcasting approach combining radar imagery with weather foundation model predictions via spectral fusion.
3/24/2026
Show Me What You Don't Know: Efficient Sampling from Invariant Sets for Model Validation
Method for analyzing feature invariances in ML models by sampling from learned equivalence classes without dedicated generators.
3/24/2026
CoRA: Boosting Time Series Foundation Models for Multivariate Forecasting through Correlation-aware Adapter
Lightweight adapter module enhancing time series foundation models by incorporating correlation information across channels.
3/24/2026
Deriving Health Metrics from the Photoplethysmogram: Benchmarks and Insights from MIMIC-III-Ext-PPG
Benchmark dataset and baselines for PPG-based clinical prediction tasks from MIMIC-III data.
3/24/2026
On the Number of Conditional Independence Tests in Constraint-based Causal Discovery
Analysis of computational complexity in constraint-based causal discovery algorithms using conditional independence tests.