6/1/2026
Unlearning's Blind Spots: Over-Unlearning and Prototypical Relearning Attack
Analysis of machine unlearning vulnerabilities including over-unlearning and relearning attacks, with metrics for class-level unlearning quality.
6/1/2026
Feature-Aware (Hyper)graph Generation via Next-Scale Prediction
FAHNES: Hierarchical framework for feature-aware hypergraph generation addressing scalability and feature preservation in graph generation.
6/1/2026
SHIELD: Secure Hypernetworks for Incremental Expansion Learning Defense
SHIELD: Certifiably robust continual learning framework combining interval bound propagation with hypernetworks for sequential task learning.
6/1/2026
Unfolding Generative Flows with Koopman Operators: Trajectory-Preserving Linearization
Method using Koopman operators to linearize and accelerate continuous normalizing flows while preserving trajectory structure.
6/1/2026
Cost-aware Stopping for Bayesian Optimization
Cost-aware stopping rules for Bayesian optimization balancing solution quality against cumulative evaluation cost.
6/1/2026
Native Hierarchical and Compositional Representations with Subspace Embeddings
Novel embedding representation using linear subspaces instead of vectors to capture hierarchies and compositional relationships in data.
6/1/2026
Beyond Tokens: Enhancing RTL Quality Estimation via Structural Graph Learning
Structural graph learning approach for RTL quality estimation in EDA workflows using graph representations beyond LLM token embeddings.
6/1/2026
Delayed Momentum Aggregation: Communication-efficient Byzantine-robust Federated Learning with Partial Participation
Byzantine-robust federated learning method with delayed momentum aggregation handling partial client participation and adversarial attacks.
6/1/2026
LiMuon: Light and Fast Muon Optimizer for Large Models
LiMuon: Lightweight variant of Muon optimizer designed for efficient training of large language models with reduced sample complexity.
6/1/2026
Towards Foundation Models for Zero-Shot Time Series Anomaly Detection: Leveraging Synthetic Data and Relative Context Discrepancy
TimeRCD: Foundation model for zero-shot time series anomaly detection using synthetic data and relative context discrepancy scoring.
6/1/2026
Spectral Collapse Drives Loss of Plasticity in Deep Continual Learning
Analysis of loss of plasticity in continual learning caused by Hessian spectral collapse, with theoretical conditions for successful training.
6/1/2026
Federated Learning with Enhanced Privacy via Model Splitting and Random Client Participation
MS-PAFL framework combining model splitting with privacy amplification for federated learning while maintaining accuracy under differential privacy.
6/1/2026
Learning a Zeroth-Order Optimizer for Fine-Tuning LLMs
ZO-Finetuner: Learning-based zeroth-order optimizer for efficient LLM fine-tuning that avoids backpropagation and reduces memory overhead.
6/1/2026
Re-examining Low Rank adaptation for private LLM fine-tuning
Analysis of low-rank adaptation (LoRA) for private LLM fine-tuning with differential privacy guarantees using DP-SGD.
6/1/2026
Adaptive Node Feature Selection For Graph Neural Networks
Method for adaptive feature selection in graph neural networks that identifies and removes unnecessary features during training.
6/1/2026
Technical note on Sequential Test-Time Adaptation via Martingale-Driven Fisher Prompting
Theoretical framework for detecting distribution shifts in streaming data using martingale-based statistical methods and Fisher information.
6/1/2026
Expand Neurons, Not Parameters
Research on increasing network neurons while keeping parameters constant to reduce feature interference and polysemanticity in neural networks.
6/1/2026
PAC-Bayesian Reinforcement Learning Trains Generalizable Policies
PAC-Bayesian generalization bound for RL accounting for Markov dependencies via chain mixing time, addressing non-independence in sequential data.
6/1/2026
Boundary-Guided Policy Optimization for Memory-efficient RL of Diffusion Large Language Models
Memory-efficient RL approach for diffusion LLMs using boundary-guided policy optimization without intractable likelihood computations.
6/1/2026
Reinforced sequential Monte Carlo for amortised sampling
Combines sequential Monte Carlo sampling with maximum-entropy RL to learn proposal kernels for sampling from unnormalized distributions.
6/1/2026
Scaling Multi-Agent Environment Co-Design with Diffusion Models
Diffusion model approach for scaling multi-agent environment co-design by jointly optimizing agent policies and environment configurations.
6/1/2026
Go-UT-Bench: A Fine-Tuning Dataset for LLM-Based Unit Test Generation in Go
Go-UT-Bench: fine-tuning dataset with 5264 Go code-test pairs for training LLMs on unit test generation tasks.
6/1/2026
Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference Mismatch
Deterministic inference method for LLMs across variable tensor parallel sizes to eliminate training-inference mismatch in agents and RL.
6/1/2026
IntAttention: A Fully Integer Attention Pipeline for Efficient Edge Inference
Integer-only attention mechanism for efficient Transformer inference on edge devices by eliminating softmax dequantization overhead.
6/1/2026
View Space: Learning Representation across Arbitrary Graphs
Learning representation framework for neural networks across arbitrary graphs with varying feature dimensionality.
6/1/2026
Bottom-up Policy Optimization: Your Language Model Policy Secretly Contains Internal Policies
RL approach decomposing LLM policies into internal layer and modular policies via Transformer residual streams for policy optimization.
6/1/2026
FEM-Bench: A Structured Scientific Reasoning Benchmark for Evaluating Code-Generating LLMs
FEM-Bench: structured benchmark for evaluating code-generating LLMs on computational mechanics and scientific reasoning tasks.
6/1/2026
Sequential Subspace Noise Injection Prevents Accuracy Collapse in Certified Unlearning
Differential privacy method for certified machine unlearning using sequential noise injection across orthogonal subspaces.
6/1/2026
Weights to Code: Extracting Interpretable Algorithms from the Discrete Transformer
Algorithm extraction technique to synthesize executable programs from Transformer weights trained on algorithmic tasks without human-written target programs.
6/1/2026
Post-Training Neural Network Pruning using Graph Curvature
Neural network pruning method using graph curvature theory to identify important connections and data flows.
6/1/2026
From Internal Diagnosis to External Auditing: A VLM-Driven Paradigm for Data-Free Online Backdoor Defense
VLM-based defense against backdoor attacks in neural networks using external semantic auditing instead of internal diagnosis.
6/1/2026
To Grok Grokking: Provable Grokking in Ridge Regression
Provable analysis of grokking phenomenon in ridge regression, demonstrating stages of overfitting and delayed generalization with gradient descent.
6/1/2026
Gap-K%: Measuring Top-1 Prediction Gap for Detecting Pretraining Data
Proposes Gap-K% method for detecting pretraining data in LLMs using prediction gap analysis, addressing privacy and copyright concerns.
6/1/2026
When More Data Doesn't Help: Limits of Adaptation in Multitask Learning
Theoretical study of multitask learning limitations, analyzing when additional data fails to improve performance in heterogeneous task settings.
6/1/2026
Don't be so Stief! Learning KV Cache low-rank approximation over the Stiefel manifold
Develops Stiefel manifold-based low-rank KV cache compression for LLMs reducing HBM capacity and bandwidth requirements in long-context decoding.
6/1/2026
Learning-to-Defer in Non-Stationary Time Series via Switching State-Space Models
Develops L2D-SLDS framework for online learning-to-defer in non-stationary time series using switching linear-Gaussian state-space models.
6/1/2026
Float8@2bits: Entropy Coding Enables Data-Free Model Compression
Proposes entropy coding compression achieving 2-bit quantization for data-free model compression without functional collapse.
6/1/2026
Scalable Topology-Preserving Graph Coarsening: Concepts and Algorithms
Introduces STPGC for topology-preserving graph coarsening with polynomial time complexity maintaining GNN performance on reduced graphs.
6/1/2026
dgMARK: Decoding-Guided Watermarking for Diffusion Language Models
Proposes dgMARK decoding-guided watermarking for discrete diffusion language models exploiting unmasking order sensitivity for output protection.
6/1/2026
SALAAD: Sparse And Low-Rank Adaptation via ADMM for Large Language Model Inference
Introduces SALAAD plug-and-play framework using sparse and low-rank adaptation via ADMM for constraining LLM inference compute and memory.
6/1/2026
The Gaussian-Head OFL Family: One-Shot Federated Learning from Client Global Statistics
Proposes Gaussian-Head OFL family for one-shot federated learning reducing communication to single round with improved privacy and deployability.
6/1/2026
Diving into Kronecker Adapters: Component Design Matters
Studies component structure design in Kronecker adapters for fine-tuning large models, exploring dimensions and quantity optimization.
6/1/2026
An Odd Estimator for Shapley Values
Analyzes theoretical mechanisms behind paired sampling heuristic for Shapley value approximation in feature importance and data valuation.
6/1/2026
Plain Transformers are Surprisingly Powerful Link Predictors
Demonstrates plain Transformers can effectively solve graph link prediction without explicit structural heuristics, offering scalable alternative to GNNs.
6/1/2026
Towards Long-Horizon Interpretability: Efficient and Faithful Multi-Token Attribution for Reasoning LLMs
Proposes efficient token attribution method for reasoning LLMs addressing O(M*N) complexity bottleneck and faithfulness challenges in long-horizon interpretability.
6/1/2026
FlexRank: Nested Low-Rank Knowledge Decomposition for Adaptive Model Deployment
Introduces FlexRank for extracting nested low-rank components from LLMs and vision transformers enabling adaptive deployment across cost budgets.
6/1/2026
Causal Evaluation of Membership Inference Attacks
Analyzes statistical validity of membership inference attacks for quantifying memorization and privacy risks, proposing causal evaluation methods.
6/1/2026
Mixture of Concept Bottleneck Experts
Proposes Mixture of Concept Bottleneck Experts framework generalizing CBMs along multiple dimensions to improve interpretability and predictive accuracy.
6/1/2026
Sequential Group Composition: A Window into the Mechanics of Deep Learning
Studies how neural networks learn structured operations through sequential group composition task, analyzing mechanisms of arithmetic and algorithmic computation.
6/1/2026
The Illusion of Generalization in Tabular Language Models
Re-evaluation of Tabular Language Models on 165 datasets revealing minimal lift over baselines and questioning claimed generalization capabilities.