4/10/2026
Inside-Out: Measuring Generalization in Vision Transformers Through Inner Workings
Method for assessing model generalization in Vision Transformers via internal representations under distribution shift.
4/10/2026
An Illusion of Unlearning? Assessing Machine Unlearning Through Internal Representations
Examines vulnerabilities in machine unlearning methods by analyzing internal representations and concept reintroduction.
4/10/2026
DMax: Aggressive Parallel Decoding for dLLMs
DMax enables efficient parallel decoding in diffusion language models through progressive self-refinement.
4/10/2026
Dead Weights, Live Signals: Feedforward Graphs of Frozen Language Models
Architecture combining frozen LLMs as nodes communicating through learned projections in a shared latent space.
4/10/2026
Leveraging Complementary Embeddings for Replay Selection in Continual Learning with Small Buffers
Continual learning method using complementary self-supervised embeddings to improve replay buffer sample selection.
4/10/2026
SOLAR: Communication-Efficient Model Adaptation via Subspace-Oriented Latent Adapter Reparametrization
Parameter-efficient fine-tuning compression framework reducing communication costs for model adaptation.
4/10/2026
ADAPTive Input Training for Many-to-One Pre-Training on Time-Series Classification
Self-supervised pre-training method for time series classification with adaptive input handling.
4/10/2026
Adversarial Label Invariant Graph Data Augmentations for Out-of-Distribution Generalization
Data augmentation method using adversarial training for out-of-distribution generalization on graphs.
4/10/2026
KV Cache Offloading for Context-Intensive Tasks
KV cache offloading technique to reduce memory and latency overhead for long-context LLM inference.
4/10/2026
Less Approximates More: Harmonizing Performance and Confidence Faithfulness via Hybrid Post-Training for High-Stakes Tasks
Hybrid post-training combining reinforcement learning and distillation to improve LLM confidence calibration.
4/10/2026
TTVS: Boosting Self-Exploring Reinforcement Learning via Test-time Variational Synthesis
Test-time variational synthesis method for reinforcement learning in domains without verifiable rewards.
4/10/2026
Quantization Impact on the Accuracy and Communication Efficiency Trade-off in Federated Learning for Aerospace Predictive Maintenance
Impact of quantization on federated learning accuracy-efficiency trade-offs for aerospace predictive maintenance.
4/10/2026
The Impact of Dimensionality on the Stability of Node Embeddings
Analysis of how embedding dimensionality affects stability of graph node embeddings.
4/10/2026
What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal
Mechanistic study of how steering vectors modify LLM behavior for alignment and refusal control.
4/10/2026
ReCodeAgent: A Multi-Agent Workflow for Language-agnostic Translation and Validation of Large-scale Repositories
Multi-agent system for language-agnostic code translation and validation across programming languages.
4/10/2026
Position Paper: From Edge AI to Adaptive Edge AI
Framework for adaptive edge AI systems that adjust models during deployment as conditions change.
4/10/2026
SHIELD: A Segmented Hierarchical Memory Architecture for Energy-Efficient LLM Inference on Edge NPUs
Memory architecture for efficient LLM inference on edge NPUs with optimized DRAM refresh for KV caches.
4/10/2026
FORGE:Fine-grained Multimodal Evaluation for Manufacturing Scenarios
Benchmark dataset and evaluation for multimodal LLMs in manufacturing scenarios.
4/10/2026
Dual-Rerank: Fusing Causality and Utility for Industrial Generative Reranking
Industrial generative reranking system combining causality and utility for video search at scale.
4/10/2026
OpenPRC: A Unified Open-Source Framework for Physics-to-Task Evaluation in Physical Reservoir Computing
Open-source framework for evaluating physical reservoir computing systems across various substrates.
4/10/2026
Munkres' General Topology Autoformalized in Isabelle/HOL
LLM-based coding agents formalized 85K lines of topology proofs in Isabelle/HOL using ChatGPT and Claude.
4/10/2026
ConsistRM: Improving Generative Reward Models via Consistency-Aware Self-Training
Paper on generative reward models for LLM alignment using consistency-aware self-training to improve scalability.
4/10/2026
Lecture notes on Machine Learning applications for global fits
Semi-autonomous multi-agent system for small molecule drug discovery using multi-modal AI agents and GNNs trained on 800M molecules.
4/10/2026
From LLM to Silicon: RL-Driven ASIC Architecture Exploration for On-Device AI Inference
RL-driven compiler using Soft Actor-Critic to jointly optimize ASIC architecture, memory hierarchy, and workload partitioning for on-device AI inference across technology nodes.
4/10/2026
Reasoning-Based Refinement of Unsupervised Text Clusters with LLMs
Framework using LLMs as semantic judges to validate and restructure outputs from unsupervised text clustering methods, improving coherence and grounding without labeled data.
4/10/2026
CAMO: A Class-Aware Minority-Optimized Ensemble for Robust Language Model Evaluation on Imbalanced Data
CAMO is an ensemble technique for imbalanced text classification that optimizes minority class performance through hierarchical voting, confidence calibration, and uncertainty estimation.
4/10/2026
From Ground Truth to Measurement: A Statistical Framework for Human Labeling
Framework for understanding systematic variation in human-labeled training data, distinguishing between ambiguous items, divergent interpretations, and mistakes rather than treating all disagreement as noise.
4/10/2026
Blink: CPU-Free LLM Inference by Delegating the Serving Stack to GPU and SmartNIC
Blink is an LLM serving architecture that removes the host CPU from the critical path by delegating orchestration and token control to GPU and SmartNIC, improving inference performance and datacenter resource utilization.
4/10/2026
DIVERSED: Relaxed Speculative Decoding via Dynamic Ensemble Verification
DIVERSED: relaxed speculative decoding for LLM inference using dynamic ensemble verification to improve token acceptance rates.
4/10/2026
IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures
IatroBench: pre-registered study documenting how AI safety measures can cause harmful model behavior changes in medical advice contexts.
4/10/2026
Efficient Dataset Selection for Continual Adaptation of Generative Recommenders
Dataset selection strategies for continual adaptation of generative recommenders under temporal distributional drift.
4/10/2026
Mitigating Distribution Sharpening in Math RLVR via Distribution-Aligned Hint Synthesis and Backward Hint Annealing
Methods to mitigate distribution sharpening in math RLVR through hint synthesis and annealing strategies.
4/10/2026
Symbiotic-MoE: Unlocking the Synergy between Generation and Understanding
Symbiotic-MoE: unified pre-training framework enabling Large Multimodal Models to generate images while maintaining understanding capabilities.
4/10/2026
Sensitivity-Positional Co-Localization in GQA Transformers
LSLoRA: investigation of sensitivity-positional co-localization in GQA transformers, restricting LoRA to optimally sensitive layers.
4/10/2026
Toward Generalizable Graph Learning for 3D Engineering AI: Explainable Workflows for CAE Mode Shape Classification and CFD Field Prediction
Graph learning framework for 3D engineering AI applications including CAE and CFD predictions with explainability.
4/10/2026
SEARL: Joint Optimization of Policy and Tool Graph Memory for Self-Evolving Agents
SEARL: framework for self-evolving AI agents that jointly optimize policy and tool graphs to learn from trajectories without large-scale LLMs.
4/10/2026
GRASS: Gradient-based Adaptive Layer-wise Importance Sampling for Memory-efficient Large Language Model Fine-tuning
GRASS: gradient-based method for memory-efficient LLM fine-tuning using adaptive layer-wise importance sampling, balancing efficiency with model expressiveness.
4/10/2026
Automatic Generation of Executable BPMN Models from Medical Guidelines
Pipeline converting healthcare policy documents to executable BPMN models using LLMs for policy simulation and evaluation.
4/10/2026
Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers
Recurrent-depth transformers enabling iterative reasoning to improve multi-hop knowledge composition in language models.
4/10/2026
Visual Perceptual to Conceptual First-Order Rule Learning Networks
Learns first-order rules from image data without labels, automatically inventing predicates for explainable AI and enhancing LLM reasoning.
4/10/2026
Dynamic Attentional Context Scoping: Agent-Triggered Focus Sessions for Isolated Per-Agent Steering in Multi-Agent LLM Orchestration
DACS mechanism for multi-agent LLM orchestration isolating per-agent context via registry and focus session modes to prevent context pollution.
4/10/2026
Generative 3D Gaussian Splatting for Arbitrary-ResolutionAtmospheric Downscaling and Forecasting
GSSA-ViT framework using 3D Gaussian splatting for arbitrary-resolution weather forecasting and downscaling of atmospheric fields.
4/10/2026
Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning
Unified framework viewing LLM post-training methods (SFT, preference optimization, RL, distillation) through off-policy and on-policy learning perspectives.
4/10/2026
Rethinking Data Mixing from the Perspective of Large Language Models
Studies data mixing strategies for LLM training, questioning domain definitions, human-model alignment, and impact of domain weighting on generalization.
4/10/2026
Are we still able to recognize pearls? Machine-driven peer review and the risk to creativity: An explainable RAG-XAI detection framework with markers extraction
Discusses risks of LLM-generated peer reviews and automated editorial processes, proposing RAG-XAI detection framework for identifying machine-generated content.
4/10/2026
DSCA: Dynamic Subspace Concept Alignment for Lifelong VLM Editing
DSCA method for lifelong vision-language model editing via dynamic subspace concept alignment, preventing degradation from sequential edits.
4/10/2026
A Decomposition Perspective to Long-context Reasoning for LLMs
Decomposes long-context reasoning in LLMs into atomic skills, automatically identifying and improving fundamental capabilities for complex reasoning.
4/10/2026
Wiring the 'Why': A Unified Taxonomy and Survey of Abductive Reasoning in LLMs
First comprehensive survey of abductive reasoning in LLMs, defining taxonomy and exploring inference of plausible explanations from observations.
4/10/2026
PrivFedTalk: Privacy-Aware Federated Diffusion with Identity-Stable Adapters for Personalized Talking-Head Generation
PrivFedTalk privacy-aware federated framework for personalized talking-head generation using diffusion models with identity-stable adapters.
4/10/2026
LINE: LLM-based Iterative Neuron Explanations for Vision Models
LINE uses LLMs iteratively to explain individual neuron concepts in vision models without predefined vocabularies, enabling interpretability of neural networks.