Contrastive Discovery scores competing explanations for scientific agents
Contrast sets resolve 73% of large belief-shift rejections to a specific alternative and resolve 67% more questions under the same experiment budget.
Stories per day · 30 days
2 RSS
Relatedbenchmark1ciphertext1clinical-nlp1contrastive-discovery1cryptography1entropy1ieee-7541language-models1llm-agents1scientific-discovery1stl1vaccine-reports1
5 stories, newest first · one card per story
Stories tagged #hypothesis-testing
STRATA scores each step of temporal clinical hypothesis tests
The benchmark has 8,735 vaccine-report hypotheses; ART narrows a record-selection gap and brings a 31B model near frontier scores.
Global and Local Entropy Tests for STL Files Under IEEE-754 Validity Constraints
Researchers derive STL-specific entropy thresholds so valid ciphertexts are not falsely rejected.
Poster: A Preliminary Study of LLM Distillation Inference
Distillation-inference test flags Llama-3.2-3B models distilled from Qwen2.5-7B at TPR 1.0 and p=0.02.
Contraction and Statistical Inference under Privacy for Uniformly Bounded Distributions
Researchers study c-interior pointwise maximal leakage for tighter privacy contraction and private statistical inference.