All Articles
19262 articles total
Rethinking Global Average Pooling: Your Classifier Is Secretly a Multi-Instance Learner
A re-evaluation of Global Average Pooling in image classifiers, proposing a Multiple-Instance Learning interpretation to recover spatial class evidence.
Regional Climate Model Emulation with Diffusion Approaches: What is the Added Value of Generative Machine Learning?
An evaluation of diffusion-based generative models for regional climate model emulation, introducing the ParamDiffusion framework.
SIMMER: Benchmarking Latent Failures in LLM Executable Planning with a World Model
SIMMER, a new benchmark for detecting latent failures in LLM-based executable planning for autonomous agents in kitchen environments.
CARE: Controlling LLM-Generated Policies through Auditable Review of Evidence in Scientific Experimentation
CARE, an auditable controller for high-throughput experimentation that uses LLMs to revise policies while maintaining a safe default optimizer.
Sensitivity Shaping for Latent Modeling
A method for improving OOD detection in robotic generative dynamics models using support-conditioned control-sensitivity regularization.
Swedish parliament abolishes permanent residence visas for migrants
Sweden's parliament has abolished permanent residence visas for migrants, marking a significant shift in immigration policy.
Why I Email Complete Strangers
An essay exploring the philosophy and practical benefits of emailing strangers to build professional and personal networks.
Reviving an abandoned open-source project: 6 years of Atomic Calendar Revive
A retrospective on the six-year journey of reviving and maintaining an abandoned open-source calendar project.
Commander Keen Games (free book)
A free book providing technical or historical insights into the Commander Keen games.
CADET: Physics-Grounded Causal Auditing and Training-Free Deconfounding of End-to-End Driving Planners
Introduction of CADET, a training-free framework to audit and repair causal confusion in end-to-end autonomous driving planners.
tap: A File-Based Protocol for Heterogeneous LLM Agent Collaboration
Presentation of tap, a file-based protocol enabling heterogeneous LLM agents from different vendors to collaborate on shared codebases.
MoDiCoL: A Modular Diagnostic Continual Learning Dataset for Robust Speech Recognition
MoDiCoL, a modular diagnostic dataset for continual learning in speech recognition, designed to improve robustness against real-world shifts.
The Perceived Fragility of Explanations in Audio Models: Manipulation of Attribution with Unchanged Predictions
Research demonstrating that explanations in audio deepfake detection models can be manipulated via psychoacoustic perturbations without changing predictions.
A Fixed-Point Neural Operator for Size- and Functional-Transferable Hamiltonian Prediction
HamEvo, a neural operator for predicting the Kohn-Sham Hamiltonian, offering significantly faster inference than conventional DFT.
Fodor and Pylyshyn's Systematicity Challenge Still Stands
A critical analysis arguing that the challenge of systematicity in human cognition remains unsolved by current neural network architectures.
Techno-libertarians are flocking to the Caribbean
A discussion regarding techno-libertarians moving to the Caribbean to establish sovereign zones or escape government regulation.
The Dead Economy Theory
An exploration of the 'Dead Economy Theory', likely discussing economic stagnation or systemic failure through a theoretical lens.
What job interviews taught me about Kubernetes
A personal reflection on the nature of Kubernetes job interviews and what they reveal about the industry's perception of the tool.
The US government’s Anthropic models ban was never about an AI jailbreak
An analysis of the US government's ban on specific Anthropic AI models, highlighting government interference in the AI industry.
PLAIground: SLO-Driven Runtime Model Selection for Compound AI Systems in the Edge-Cloud-Space Continuum
Introduces PLAIground, a framework for SLO-driven runtime model selection in Compound AI systems across edge, cloud, and space environments.