The Archive

Initially launched in 2019, the Nvidia Shield TV Pro now retails for $299.99.

Clay Co-founder and CEO Kareem Amin joins the AI Stage to discuss the rise of GTM engineer at TechCrunch Disrupt 2026. Register for your ticket and get a second pass at 50% off.

Ahmad Al-Dahle, ex-Meta Llama lead, applies AI to Airbnb's product development and guest experience.

S²-PINN applies separable neural networks to uncertainty quantification in random PDEs using Gaussian dictionaries and Fourier features.

JOVE assigns heterogeneous LLMs to task graph subtasks with asynchronous verification to optimize latency and cost trade-offs.

Transfer learning from echocardiography pretraining to lung ultrasound for TB screening; V-JEPA2-L encoder shows strong domain cross-over.

Mechanistic analysis of vision-language fusion in MLLMs via attention routing, alignment decoupling, and dimensionality reduction across architectures.

EVOL uses simulator-guided evolutionary synthesis to generate synthetic expert learning paths for sparse-reward RL in educational recommendation.

Soft-vote ensemble and counter-causal augmentation for Touché 2026 causality extraction task on news sentences with polarity classification.

SPEAR applies spectral-disentangled mixture-of-experts to PDE foundation models to mitigate knowledge interference and expert redundancy.

Consecutive Posterior Fusion extends diffusion posterior sampling by reusing prior measurement corrections for ill-posed inverse imaging problems.

PaMIR: open benchmark of 19 public credit-default datasets (1.24M accounts across 9 countries) with late-arriving labels and leakage auditing.

Empirical comparison of four nonlinear causal discovery approaches reveals accuracy-runtime trade-offs; differentiable/amortized methods scale but lose accuracy.

ML models (LR, MLP, CNN) evaluated on CAGE-TB dataset for TB screening across South Africa and Uganda cohorts.

Models distinguish fact memorization from retrieval sensitivity; identical statements retrieved better than paraphrased requests across training regimes.

Collective Bias Mitigation framework uses multi-model routing to reduce stereotypes in LLM outputs deployed in healthcare and governance.

WAMpy synthesizes Prolog programs via NumPy-compiled WAM instructions and JIT acceleration for repeated hypothesis evaluation.

D2K-Bench diagnostic evaluates LLM agent GPU kernel generation quality across algorithmic, dataflow, and low-level optimization guidance levels.

Neuro-Physical Inverter couples ensemble conditioning with residual neural networks for 1D magnetotelluric geophysical inversion.

AortaDiff diffusion model generates contrast-enhanced CT; model uncertainty correlates with anatomical correctness in medical image synthesis.

AdaStep adaptively weights step-level credit assignments for long-horizon LLM agent RL training, balancing sparse rewards with trajectory noise.

Convex optimization complexity bounds with lazy second-order oracles: lower bound Ω(m + m^1/7 ε^-2/7) and matching upper bound algorithm.

Neural architecture search optimizes parameterized quantum circuit design integrated with classical deep learning for image classification tasks.

Extended kernel SVD method for joint nonlinear feature learning across multiple asymmetric data sources.

StanceEval 2026 shared task: stance detection in Arabic social media with cross-target generalization tracks.

SLM-based intent matching for tool-call oversight in agents, enabling low-latency semantic validation of action relevance.

HyperFuse: fast self-supervised node embedding method for attributed hypergraphs via spectral relaxation.

Quantum complexity bounds for Hamiltonian property testing without inverse evolution operator access.

COFLOW: adaptive step-count selection in flow models for efficient image generation via context-aware inference.

PrO-GPs: Gaussian Process variant prioritizing predictive uncertainty over standard kernel/model specification.