4/1/2026
SAM3-CPU – Run Segment Anything on CPU with memory-aware video chunking
CPU-compatible wrapper for Meta's Segment Anything Model 3 with memory-aware video chunking and intelligent GPU/CPU device selection.
4/1/2026
CockroachDB has just dropped a bunch of AI-ready features
CockroachDB releases agent-ready features including managed MCP Server and Agent Skills library for AI agents database integration.
4/1/2026
Show HN: Max Headbox, a local agent that fits on a Raspberry Pi 5
Max Headbox is an open-source voice-activated LLM agent running 100% locally on Raspberry Pi 5, supporting custom tools and local automation without cloud dependency.
4/1/2026
Airupt – open-source red-teaming for LLMs (79 attack vectors)
Airupt open-source tool for red-teaming LLMs with 79 attack vectors. Security and robustness testing framework.
4/1/2026
Mesh – one room for your AI agents (open source)
Mesh is an open-source real-time chat room enabling multiple AI agents (Claude, Cursor, Gemini) to collaborate, communicate, and hand off tasks via MCP protocol.
4/1/2026
Show HN: Castra – Strip orchestration rights from your LLMs
Castra: Go-based orchestration layer separating LLM state from chat history using encrypted SQLite. Addresses context window drift in agents.
4/1/2026
KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao
KARMA framework addresses knowledge-action gap when fine-tuning LLMs for personalized search tasks at scale, proposing regularization methods for industrial recommendation systems.
4/1/2026
Robust Safety Monitoring of Language Models via Activation Watermarking
Activation watermarking technique for detecting adversarial attacks on LLMs that evade safety monitoring while eliciting unsafe outputs.
4/1/2026
Cost-Sensitive Neighborhood Aggregation for Heterophilous Graphs: When Does Per-Edge Routing Help?
GNN layer with per-edge routing for heterophilous graphs, comparing cost-sensitive aggregation against uniform spectral approaches.
4/1/2026
SleepVLM: Explainable and Rule-Grounded Sleep Staging via a Vision-Language Model
Vision-language model for sleep staging from polysomnography waveforms generating AASM-compliant clinical rationales with auditable reasoning.
4/1/2026
Magic Words or Methodical Work? Challenging Conventional Wisdom in LLM-Based Political Text Annotation
Controlled evaluation of how LLM model choice, size, and prompting strategies affect political text annotation, challenging conventional wisdom.
4/1/2026
Cross-attentive Cohesive Subgraph Embedding to Mitigate Oversquashing in GNNs
GNN technique using cross-attention and cohesive subgraph embedding to address oversquashing problem in graph neural networks.
4/1/2026
ITQ3_S: High-Fidelity 3-bit LLM Inference via Interleaved Ternary Quantization with Rotation-Domain Smoothing
3-bit weight quantization method for LLMs using rotation-domain smoothing via Fast Walsh-Hadamard Transform, improving precision in extreme quantization.
4/1/2026
JaWildText: A Benchmark for Vision-Language Models on Japanese Scene Text Understanding
Benchmark dataset for evaluating vision-language models on Japanese scene text understanding, addressing multilingual complexity challenges.
4/1/2026
Building evidence-based knowledge graphs from full-text literature for disease-specific biomedical reasoning
EvidenceNet framework uses LLM-assisted pipeline to extract structured, evidence-grounded biomedical findings from full-text literature into knowledge graphs.
4/1/2026
ResAdapt: Adaptive Resolution for Efficient Multimodal Reasoning
Adaptive resolution framework for multimodal LLMs that reduces visual token overhead through input-side compression before encoding.
4/1/2026
OneComp: One-Line Revolution for Generative AI Model Compression
Framework for post-training compression of generative AI models with single-line implementation, addressing quantization and calibration challenges.
4/1/2026
Beta-Scheduling: Momentum from Critical Damping as a Diagnostic and Correction Tool for Neural Network Training
Derives time-varying momentum schedule for neural network training from physics principles, eliminating need for manual tuning.
4/1/2026
Differentiable Initialization-Accelerated CPU-GPU Hybrid Combinatorial Scheduling
Hybrid CPU-GPU framework combining differentiable optimization with ILP solving for combinatorial scheduling.
4/1/2026
Multi-Agent LLMs for Adaptive Acquisition in Bayesian Optimization
Multi-agent LLM framework for Bayesian optimization exploring exploration-exploitation trade-off through implicit reasoning.
4/1/2026
Improving Efficiency of GPU Kernel Optimization Agents using a Domain-Specific Language and Speed-of-Light Guidance
LLM agents for GPU kernel optimization using domain-specific language and speed-of-light guidance to reduce design space.
4/1/2026
ARCS: Autoregressive Circuit Synthesis with Topology-Aware Graph Attention and Spec Conditioning
Amortized analog circuit generation system combining graph VAE and flow-matching models with SPICE validation.
4/1/2026
Realistic Market Impact Modeling for Reinforcement Learning Trading Environments
Gymnasium-compatible RL trading environments with realistic nonlinear market impact models for agent evaluation.
4/1/2026
HCLSM: Hierarchical Causal Latent State Machines for Object-Centric World Modeling
Hierarchical world model with object-centric decomposition and causal latent dynamics for video prediction.
4/1/2026
Efficient Bilevel Optimization with KFAC-Based Hypergradients
Bilevel optimization using KFAC-based hypergradients for efficient inverse Hessian-vector product computation.
4/1/2026
Efficient and Scalable Granular-ball Graph Coarsening Method for Large-scale Graph Node Classification
Graph coarsening method for scalable Graph Convolutional Networks on large-scale node classification tasks.
4/1/2026
Deep Learning-Based Anomaly Detection in Spacecraft Telemetry on Edge Devices
Three deep learning approaches for spacecraft telemetry anomaly detection optimized for edge device deployment using neural architecture search.
4/1/2026
mtslearn: Machine Learning in Python for Medical Time Series
Open-source Python library for machine learning on medical time-series data, addressing heterogeneous clinical data and reducing friction for ML practitioners in healthcare applications.
4/1/2026
An Isotropic Approach to Efficient Uncertainty Quantification with Gradient Norms
Lightweight uncertainty quantification for neural networks using gradient norms and isotropy assumption without training data access.
4/1/2026
Survival In-Context: Prior-fitted In-context Learning Tabular Foundation Model for Survival Analysis
Prior-fitted tabular foundation model using in-context learning for survival analysis with limited and censored data.
4/1/2026
Why not to use Cosine Similarity between Label Representations
Analysis showing cosine similarity between label representations in softmax classifiers does not reliably indicate model behavior.
4/1/2026
Target-Aligned Reinforcement Learning
Target-Aligned RL (TARL) framework addressing stability-recency tradeoff in target networks through selective emphasis of aligned transitions.
4/1/2026
Disentangled Graph Prompting for Out-Of-Distribution Detection
Graph prompt-based method for out-of-distribution detection in neural networks using disentangled representations.
4/1/2026
A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models
Information decomposition framework measuring information spectrum in vision-language models to assess multimodal fusion vs unimodal priors.
4/1/2026
Mind the Gap: A Framework for Assessing Pitfalls in Multimodal Active Learning
Framework analyzing pitfalls in active learning for multimodal data, addressing missing modalities and varying interaction structures.
4/1/2026
One-for-All: A Lightweight Stabilized and Parameter-Efficient Pre-trained LLM for Time Series Forecasting
One-for-All: parameter-efficient LoRA variant (rsLoRA) for adapting frozen LLMs to multivariate time-series forecasting tasks.
4/1/2026
Training-Free Dynamic Upcycling of Expert Language Models
Training-free method to combine multiple domain-specific expert LLMs into single multi-domain model without fine-tuning.
4/1/2026
Big2Small: A Unifying Neural Network Framework for Model Compression
Big2Small unifies model compression techniques (pruning, quantization, distillation, decomposition) under single mathematical framework.
4/1/2026
From Density Matrices to Phase Transitions in Deep Learning: Spectral Early Warnings and Interpretability
Uses reduced density matrices from quantum chemistry to predict phase transitions in neural networks during training and improve interpretability.
4/1/2026
Loss Gap Parity for Fairness in Heterogeneous Federated Learning
Proposes EAGLE, a federated learning algorithm ensuring fair performance across heterogeneous clients by minimizing loss gap parity.
4/1/2026
Curvature-Guided LoRA: Steering in the pretrained NTK subspace
Curvature-Guided LoRA: Parameter-efficient fine-tuning approach using prediction alignment to match full fine-tuning performance.
4/1/2026
Task Scarcity and Label Leakage in Relational Transfer Learning
Study of label leakage problem in relational transfer learning where task scarcity causes models to learn task-specific shortcuts.
4/1/2026
Real-Time Explanations for Tabular Foundation Models
ShapPFN: Foundation model integrating Shapley value regression for real-time interpretable predictions on tabular data.
4/1/2026
Meteorology-Driven GPT4AP: A Multi-Task Forecasting LLM for Atmospheric Air Pollution in Data-Scarce Settings
GPT4AP: Parameter-efficient multi-task forecasting framework using rsLoRA for air pollution prediction in data-scarce regions.
4/1/2026
Aligning Validation with Deployment: Target-Weighted Cross-Validation for Spatial Prediction
Target-Weighted Cross-Validation method for improving predictive risk estimation in spatial prediction with structured data.
4/1/2026
Tracking Equivalent Mechanistic Interpretations Across Neural Networks
Framework for discovering and validating mechanistic interpretations across neural networks to improve interpretability and generalization.
4/1/2026
Tucker Attention: A generalization of approximate attention mechanisms
Research on Tucker attention as a generalization of approximate attention mechanisms like GQA and MLA using low-rank factorizations.
4/1/2026
Reward-Based Online LLM Routing via NeuralUCB
NeuralUCB-based cost-aware LLM routing algorithm that adapts online to model performance and cost, outperforming supervised routing baselines.
4/1/2026
CRAFT: Cost-aware Expert Replica Allocation with Fine-Grained Layerwise Estimations
CRAFT: Cost-aware expert replica allocation for mixture-of-experts LLM serving with fine-grained layerwise load-balancing estimations.
4/1/2026
Spark-LLM-Eval: A Distributed Framework for Statistically Rigorous Large Language Model Evaluation
Spark-LLM-Eval: Distributed framework for statistically rigorous evaluation of LLMs at scale across hundreds of thousands of samples.