3/23/2026
Breaking the Capability Ceiling of LLM Post-Training by Reintroducing Markov States
Research identifying capability ceiling in LLM RL post-training by reintroducing Markov state structure for improved alignment.
3/23/2026
A Super Fast K-means for Indexing Vector Embeddings
SuperKMeans, a k-means variant for clustering vector embeddings 7x faster than FAISS/Scikit-Learn on CPUs.
3/23/2026
AgenticRS-EnsNAS: Ensemble-Decoupled Self-Evolving Architecture Search
AgenticRS-EnsNAS addresses NAS validation bottleneck in ensemble-based production systems with decoupled architecture search.
3/23/2026
ODySSeI: An Open-Source End-to-End Framework for Automated Detection, Segmentation, and Severity Estimation of Lesions in Invasive Coronary Angiography Images
Open-source framework for automated detection, segmentation, and severity estimation of lesions in coronary angiography images.
3/23/2026
Continual Learning as Shared-Manifold Continuation Under Compatible Shift
Continual learning method treating representation evolution as shared-manifold continuation to prevent catastrophic forgetting.
3/23/2026
Federated Hyperdimensional Computing for Resource-Constrained Industrial IoT
Hyperdimensional computing for federated learning on resource-constrained IoT edge devices.
3/23/2026
Fine-tuning Timeseries Predictors Using Reinforcement Learning
RL algorithms for fine-tuning financial forecasters with supervised learning models, showing improved performance via transfer learning.
3/23/2026
How Out-of-Equilibrium Phase Transitions can Seed Pattern Formation in Trained Diffusion Models
Theoretical framework interpreting diffusion model generation as out-of-equilibrium phase transitions with architectural analysis.
3/23/2026
Spectral Alignment in Forward-Backward Representations via Temporal Abstraction
Research on spectral alignment in forward-backward representations for successor representation learning in continuous spaces.
3/23/2026
The $\mathbf{Y}$-Combinator for LLMs: Solving Long-Context Rot with $\lambda$-Calculus
λ-RLM: framework using λ-calculus to solve long-context problems in LLMs via recursive decomposition with verifiable control code execution.
3/23/2026
Trojan horse hunt in deep forecasting models: Insights from the European Space Agency competition
Analysis of trojan horse attacks in deep forecasting models using insights from ESA competition, focusing on backdoor triggers in safety-critical applications.
3/23/2026
Var-JEPA: A Variational Formulation of the Joint-Embedding Predictive Architecture -- Bridging Predictive and Generative Self-Supervised Learning
Var-JEPA: variational formulation bridging JEPA and probabilistic generative modeling, showing their structural equivalence beyond rhetorical separation.
3/23/2026
Conditioning Protein Generation via Hopfield Pattern Multiplicity
Method for conditioning protein sequence generation via stochastic attention using Hopfield pattern multiplicity to direct generation toward functional subsets.
3/23/2026
Revisiting Gene Ontology Knowledge Discovery with Hierarchical Feature Selection and Virtual Study Group of AI Agents
Virtual study group of AI agents for knowledge discovery in gene ontology using LLMs and agentic AI to extract aging-related biological knowledge.
3/23/2026
Beyond Single Tokens: Distilling Discrete Diffusion Models via Discrete MMD
D-MMD: discrete moment matching distillation method for distilling discrete diffusion models to reduce sampling steps while maintaining quality and diversity.
3/23/2026
Kolmogorov-Arnold causal generative models
KaCGM: causal generative model for mixed-type tabular data using Kolmogorov-Arnold Networks for interpretable structural equations and causal inference.
3/23/2026
MAPLE: Metadata Augmented Private Language Evolution
MAPLE: method for generating differentially private synthetic data from LLMs accessed via APIs, enabling DP fine-tuning without direct model access.
3/23/2026
Significance-Gain Pair Encoding for LLMs: A Statistical Alternative to Frequency-Based Subword Merging
Significance-Gain BPE: statistical alternative to frequency-based subword tokenization for LLMs that measures pair cohesion rather than raw frequency for improved tokenization.
3/23/2026
Probing to Refine: Reinforcement Distillation of LLMs via Explanatory Inversion
Novel LLM distillation framework using reinforcement distillation and explanatory inversion to transfer reasoning capabilities to smaller student models with better generalization.
3/23/2026
Autonoma: A Hierarchical Multi-Agent Framework for End-to-End Workflow Automation
Autonoma: hierarchical multi-agent framework for translating natural language instructions into robust multi-step workflows, addressing scalability and error propagation in workflow automation.
3/23/2026
MOSAIC: Modular Opinion Summarization using Aspect Identification and Clustering
Modular summarization framework decomposing review processing into interpretable components for opinion extraction and clustering.
3/23/2026
Generalized Stock Price Prediction for Multiple Stocks Combined with News Fusion
LLM-based approach combining daily financial news with stock embeddings for multi-stock price prediction.
3/23/2026
VERDICT: Verifiable Evolving Reasoning with Directive-Informed Collegial Teams for Legal Judgment Prediction
LLM-based multi-agent framework for legal judgment prediction with interpretable reasoning and evolving case analysis.
3/23/2026
Towards Solving Polynomial-Objective Integer Programming with Hypergraph Neural Networks
Hypergraph neural network method for solving polynomial-objective integer programming problems with nonlinear constraints.
3/23/2026
Exploring the Agentic Frontier of Verilog Code Generation
Study of agentic frameworks with LLMs for Verilog hardware code generation, comparing agent-based versus standard approaches.
3/23/2026
Automated Membership Inference Attacks: Discovering MIA Signal Computations using LLM Agents
Using LLM agents to automate discovery of membership inference attacks against machine learning models.
3/23/2026
Scalable Prompt Routing via Fine-Grained Latent Task Discovery
Dynamic routing system selecting optimal LLM from large model pools using fine-grained latent task discovery.
3/23/2026
Pseudo-Labeling for Unsupervised Domain Adaptation with Kernel GLMs
Principled framework for unsupervised domain adaptation in kernel GLMs under covariate shift using pseudo-labeling.
3/23/2026
The Autonomy Tax: Defense Training Breaks LLM Agents
Analysis showing safety-focused defense training degrades LLM agent tool-use capabilities, revealing capability-alignment paradox.
3/23/2026
Vocabulary shapes cross-lingual variation of word-order learnability in language models
Study on how vocabulary influences word-order learnability in transformer language models across languages.
3/23/2026
Reinforcement-guided generative protein language models enable de novo design of highly diverse AAV capsids
Protein language model with reinforcement learning guidance for designing diverse AAV capsids for gene therapy.
3/23/2026
Teaching an Agent to Sketch One Part at a Time
Multi-modal language model agent trained with process-reward RL to generate vector sketches part-by-part.
3/23/2026
ReXInTheWild: A Unified Benchmark for Medical Photograph Understanding
Benchmark for evaluating vision-language models on medical photograph understanding tasks.
3/23/2026
EvidenceRL: Reinforcing Evidence Consistency for Trustworthy Language Models
Reinforcement learning framework ensuring LLMs ground responses in evidence to reduce hallucinations.
3/23/2026
Verifiable Error Bounds for Physics-Informed Neural Network Solutions of Lyapunov and Hamilton-Jacobi-Bellman Equations
Method for verifiable error bounds in physics-informed neural networks solving Lyapunov and HJB equations.
3/23/2026
ATHENA: Adaptive Test-Time Steering for Improving Count Fidelity in Diffusion Models
Test-time adaptive steering method improving object count accuracy in text-to-image diffusion models without retraining.
3/23/2026
A Subgoal-driven Framework for Improving Long-Horizon LLM Agents
Framework improving long-horizon LLM agent planning using subgoals for complex web navigation and digital environment control tasks.
3/23/2026
Diminishing Returns in Expanding Generative Models and Godel-Tarski-Lob Limits
Theoretical analysis of scaling limits in generative models through Godel-Tarski-Lob framework examining capability growth bounds.
3/23/2026
Growing Networks with Autonomous Pruning
Neural network architecture that dynamically grows and prunes parameters during training for efficient image classification.
3/23/2026
IsoCLIP: Decomposing CLIP Projectors for Efficient Intra-modal Alignment
Analysis of CLIP vision-language model projectors to improve intra-modal alignment for image retrieval tasks.
3/23/2026
Deep Autocorrelation Modeling for Time-Series Forecasting: Progress and Prospects
Survey of deep learning architectures and objectives for modeling autocorrelation in time-series forecasting tasks.
3/23/2026
HiPath: Hierarchical Vision-Language Alignment for Structured Pathology Report Prediction
Vision-language model for structured pathology report prediction treating multi-granular diagnostic output as primary objective.
3/23/2026
Evaluating Test-Time Adaptation For Facial Expression Recognition Under Natural Cross-Dataset Distribution Shifts
Evaluation of test-time adaptation methods for facial expression recognition under natural cross-dataset distribution shifts.
3/23/2026
Layered Quantum Architecture Search for 3D Point Cloud Classification
Quantum architecture search method progressively growing parametrized quantum circuits for 3D point cloud classification tasks.
3/23/2026
Graph-Informed Adversarial Modeling: Infimal Subadditivity of Interpolative Divergences
Theoretical analysis of adversarial learning with graph-factorized distributions using interpolative divergences and variational methods.
3/23/2026
Structured Latent Dynamics in Wireless CSI via Homomorphic World Models
Self-supervised framework learning predictive wireless channel representations using JEPA world modeling on CSI trajectories.
3/23/2026
Semantic Token Clustering for Efficient Uncertainty Quantification in Large Language Models
Improves HAL semantic representations using attention-based pooling for text classification. Addresses information loss in mean pooling aggregation.
3/23/2026
Measuring Faithfulness Depends on How You Measure: Classifier Sensitivity in LLM Chain-of-Thought Evaluation
Semantic token clustering method for uncertainty quantification in LLMs without repeated sampling. Addresses computational overhead in reliability assessment.
3/23/2026
AI Agents Can Already Autonomously Perform Experimental High Energy Physics
Shows chain-of-thought faithfulness measurement varies by classifier, undermining single aggregate metrics. Evaluates 12 open-weight LLM models with 10K+ traces.
3/23/2026
From Masks to Pixels and Meaning: A New Taxonomy, Benchmark, and Metrics for VLM Image Tampering
Demonstrates Claude Code autonomously executing high energy physics analysis pipeline including event selection, background estimation, and statistical inference.