A new unsupervised framework generates pseudo-anomalies inside a latent space shaped by a pre-trained LLM, beating existing baselines on the TAB benchmark.
A new adaptive conformal prediction method learns per-prompt thresholds via quantile regression, fixing over and under filtering of LLM claims across diverse topics.
Hugging Face ran 90 controlled experiments to figure out what actually makes rephrased web text useful for pretraining, and released a 486B-token dataset built on the findings.
GLOW pairs a lightweight graph neural network with an off-the-shelf LLM to answer questions when the knowledge graph is missing facts, beating prior hybrids by up to 53 points.
A systematic reward ablation on physics problems shows attention-based supervision doubles spatial reasoning accuracy but breaks symbolic math domains.
RWTH Aachen researchers show masked and uniform-state diffusion models can rescore ASR hypotheses and even join CTC decoding, cutting word error rates.
A radar processing method that works in the frequency domain promises robust odometry and simultaneous multi-object tracking for high-speed autonomous driving.
A new fairness framework rejects prediction-based parity metrics in favor of stakeholder utilities, showing stochastic policies expand Pareto trade-offs.
Researchers turn the informal practice of vibe-testing into a structured evaluation pipeline, and show that personalized prompts can flip which model wins.
Researchers propose reinforcement learning on the marginal distribution P(y) instead of P(y|x), then combine both stages to expand reasoning exploration beyond GRPO limits.
A dual-stage framework injects syntactic dependency information directly into single-GPU BERT pretraining, adding 0.83 GLUE points at no extra compute cost.
Researchers at Meta propose a Kuiper-statistic metric weighted by signal-to-noise ratios, arguing that unweighted multi-calibration scores are mostly noise.
Researchers show 4-bit quantization silently wrecks differentially private training, and offer a dynamic layer-rotation fix that keeps accuracy within 2 percent.