6/1/2026
GEM: Geometric Entropy Mixing for Optimal LLM Data Curation
GEM reformulates LLM pre-training data curation as variational optimization on hypersphere to address ontological misalignment and embedding anisotropy in data mixing.
6/1/2026
Pair-In, Pair-Out: Latent Multi-Token Prediction for Efficient LLMs
Pair-In, Pair-Out proposes latent multi-token prediction combining input-side compression and output-side efficiency to reduce LLM inference costs.
6/1/2026
No Reader Left Behind: Multi-Agent Summaries Everyone Can Understand
NRLB is a multi-agent framework for plain language summarization that adapts outputs for diverse reader groups with different linguistic and cognitive abilities.
6/1/2026
First head-to-head comparison of agentic AI applied to the analysis of simulated data of the Einstein Telescope
Head-to-head comparison of Claude Code and Codex executing autonomous gravitational wave data analysis pipelines on shared infrastructure without human intervention.
6/1/2026
SafeRx-Agent: A Knowledge-Grounded Multi-Agent Framework for Safe and Explainable Medication Recommendation
SafeRx-Agent proposes a multi-agent framework combining LLMs with safety verification for medication recommendation, addressing explainability and traceability in clinical decision-making.
6/1/2026
Compute Allocation in Evolutionary Search: From Depth-Breadth to Multi-Armed Bandits
Studies compute allocation strategies in LLM-guided evolutionary search across depth-breadth tradeoff using multi-armed bandit approach.
6/1/2026
Pocket-Dentist: On-Device Dental Image Understanding via Efficient Multimodal Large Language Models
Pocket-Dentist: efficient multimodal LLM for on-device dental image analysis optimizing for inference speed and privacy.
6/1/2026
No More K-means: Single-Stage Sparse Coding for Efficient Multi-Vector Retrieval
Efficient sparse coding approach for multi-vector retrieval replacing k-means clustering with single-stage processing.
6/1/2026
QASM-Eval: A Dataset to Train and Evaluate LLMs on OpenQASM-3 Beyond Quantum Circuits
QASM-Eval dataset trains LLMs on OpenQASM-3 quantum programming including error correction, timing, and pulse-level operations.
6/1/2026
Unicorn: Scaling High-Dimensional Time Series Forecasting via Universal Correlation Modeling
Unicorn framework for scalable multi-dataset time series forecasting balancing channel-independence and channel-dependency modeling.
6/1/2026
When LLMs Learn to Be Consistently Wrong: A Multi-Model Study of Linear Representations of Synthetic Deception
Multi-model study of linear representations in LLM deceptive alignment using synthetic dishonesty as controlled testbed.
6/1/2026
LLMs Without Deep Neural Networks: New Architecture, Benefits and Case Study
Proposes RBF networks as alternative LLM architecture without deep neural networks claiming improved explainability.
6/1/2026
NumLeak: Public Numeric Benchmarks as Latent Labels in Foundation Models
NumLeak measurement framework detecting memorized benchmark data in frontier LLMs via API probes and white-box analysis.
6/1/2026
LongDS-Bench: On the Failure of Long-Horizon Agentic Data Analysis
LongDS-Bench evaluates long-horizon agentic data analysis with 68 real-world Kaggle-based tasks testing context tracking across multi-turn interactions.
6/1/2026
Calibrated Preference Learning: The Case of Label Ranking
ML research extending calibration theory to probabilistic label ranking prediction tasks with structured output spaces.
6/1/2026
Bounded Behavioral Indistinguishability for Black-Box LLM Distillation
Framework for evaluating LLM distillation beyond output matching using bounded behavioral indistinguishability formal measure.
6/1/2026
VeriGate: Verifier-Gated Step-Level Supervision for GRPO
VeriGate improves GRPO reasoning model training by adding step-level verifier feedback to address sparse supervision and credit assignment.
6/1/2026
A Unified Framework for Gradient Aggregation in Multi-Objective Optimization
Unified theoretical framework analyzing gradient aggregation methods in multi-objective optimization with convergence rate analysis.
6/1/2026
DisjunctiveNet: Neural Symbolic Learning via Differentiable Convexified Optimization Layers
Neuro-symbolic learning framework integrating neural networks with differentiable optimization for incorporating domain knowledge as logical rules.
6/1/2026
Scalable Constrained Multi-Agent Reinforcement Learning via State Augmentation and Consensus for Separable Dynamics
Distributed multi-agent reinforcement learning approach for constrained coordination problems with separable agent dynamics.
6/1/2026
Can Subgraph Explanations Be Weaponized to Steal Graph Neural Networks?
Model extraction attack against graph neural networks using explainability interfaces in GMLaaS platforms.
6/1/2026
Universal Multiclass Transductive Online Learning
Theoretical framework for universal multiclass transductive online learning with unbounded label spaces.
6/1/2026
Discovering a Zeta Map Algorithm on Dyck Paths via Mechanistic Interpretability
Mechanistic interpretability study recovering explicit Zeta map algorithm on Dyck paths from trained transformer.
6/1/2026
Graph-Conditioned Mixture of Graph Neural Network Experts for Traffic Forecasting
Graph-conditioned mixture of experts framework for spatio-temporal traffic forecasting with node-wise expert specialization.
6/1/2026
MAAT: Multi-phase Adapter-Aware Targeted Unlearning
Balanced benchmark and unlearning method for evaluating machine unlearning across causal and relational knowledge.
6/1/2026
Representation Collapse in Sequential Post-Training of Large Language Models
Study of representation collapse during sequential post-training of LLMs with measurement suite for hidden states and LoRA updates.
6/1/2026
Measuring, Localizing, and Ablating Alignment Signatures in LLMs
Analysis of AI-style alignment signatures in LLMs through measurement and localization of post-training effects on representations.
6/1/2026
The Long-Term Effects of Data Selection in LLM Fine-Tuning
Study of long-term effects of data selection strategies in multi-stage LLM fine-tuning and model adaptability.
6/1/2026
SubsurfaceGen: Procedural Generation of Field-Scale Earth Models and Seismic Data
Procedural generation system for creating field-scale earth models and seismic data for full waveform inversion training.
6/1/2026
Early Prediction of Future Behavioral Strategy from Process Traces
Method for early prediction of human behavioral strategy from process traces for adaptive systems.
6/1/2026
Destruction is a General Strategy to Learn Generation; Diffusion's Strength is to Take it Seriously; Exploration is the Future
Theoretical perspective on diffusion models as information-withholding techniques and their advantages in data-scarce settings.
6/1/2026
Supervised Training Rapidly Degrades Early Visual Cortex Alignment Across Biologically Plausible Learning Rules
Study showing supervised training degrades visual cortex alignment in neural networks across multiple biologically plausible learning rules.
6/1/2026
Benchmarking Machine Learning Uncertainty Quantification Methodologies for Predicting Turbine Gas Temperature Degradation
Benchmarking five uncertainty quantification methods for turbine gas temperature degradation prediction.
6/1/2026
Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents
Causal Sensitivity Score metric to evaluate hidden capabilities in clinical LLMs and agents through counterfactual case mutations.
6/1/2026
Learning Transferable Predictability Representations
Method for assigning predictability scores to trajectory windows across different dynamics regimes using ordinal estimation.
6/1/2026
Scientific Machine Learning for Engine Health Management and Remaining Useful Life Prediction
Multi-task machine learning framework for turbine engine health management and remaining useful life prediction using real-world fleet data.
6/1/2026
Improving Relative Representations with Learned Anchors and Whitened Inner Products
Research on improving relative representations in neural networks using learned anchors and whitened inner products to enable modular AI systems.
6/1/2026
AMNESIA: A Large Scale Medical Unlearning Benchmark Suite with Disease-Informed Analysis
AMNESIA: first large-scale open-source benchmark for medical machine unlearning, enabling selective forgetting in medical LLMs.
6/1/2026
TASER: Task-Aware Stein Regularisation for Geometry-Driven Robustness
TASER: training-time regularization framework using Langevin Stein operators to improve robustness under distribution shift and adversarial perturbations.
6/1/2026
Constrained Flow Optimization via Sequential Fine Tuning for Molecular Design
Fine-tuning method for adapting diffusion/flow models to optimize rewards while satisfying constraints for molecular design.
6/1/2026
CSULoRA: Closest Safe Update Low-Rank Adaptation
CSULoRA: safety-preserving low-rank adaptation method preventing adversarial fine-tuning from weakening LLM alignment.
6/1/2026
Diffusion Models Preferentially Memorize Prototypical Examples or: Why Does My Diffusion Model Love Slop?
Study showing diffusion models memorize prototypical examples rather than atypical ones. Analysis of training data memorization patterns.
6/1/2026
LARK: Learnability-Grounded Trajectory Selection for Efficient Reasoning Distillation
LARK: method for selecting learnable reasoning trajectories for efficient knowledge distillation from teacher to student models.
6/1/2026
Bridging the Gap Between Natural Language and Market Dynamics via High-Dimensional Representation Learning
Using FinBERT embeddings with Transformers for financial forecasting instead of scalar sentiment scores. Application of representation learning to finance.
6/1/2026
Learning to Perceive the World Through Control: Empowerment-Based Representation Learning
Research on learning control-relevant representations via empowerment objective for unsupervised skill learning in reinforcement learning.
6/1/2026
BOKBO (Best of K Bad Options): Calibrated Abstention for VLA Policies
BOKBO: conformal abstention layer for vision-language-action policies providing distribution-free safety guarantees on executed-violation rate.
6/1/2026
Universal Decision Learners
Categorical framework unifying decision-making theories (planning, RL, causal intervention, online learning) using universal constructions.
6/1/2026
Diversity Matters: Revisiting Test-Time Compute in Vision-Language Models
Systematic study of test-time compute strategies in vision-language models across seven models and six benchmarks using feature-based scoring and voting.
6/1/2026
When are LLMs Sufficient Policy Optimizers for Sequential RL Tasks?
Prompted Policy Optimization (PromptPO): using LLMs as black-box policy optimizers for reinforcement learning tasks with executable policy generation.
6/1/2026
Reducing the GPU Memory Bottleneck with Lossless Compression for ML -- Extended
Lossless compression techniques to reduce GPU memory bottlenecks during ML training and inference without accuracy loss.