MACHINE LEARNING

The same posture, one layer up — measuring models, not molecules.

Latent embedding space, computed constellation
Scroll to descend
Latent embedding space
HeLa cell in anaphase — the origin specimen
Epoch II · 2024 —

Models, under measurement

Jailbreaks and their mitigation. Uncertainty, from aleatoric to epistemic. Explanations that survive scrutiny. The clinical systems these models now touch.

The origin · the posture

The posture came from the bench

Measure carefully. Quantify error honestly. Never trust a black box you have not interrogated — whether it holds cells or weights.

Fig 3.1 — Latent space · embedding clusters
01

LLM security

How large models are attacked, and how they are defended: Jailbreaking and Mitigation of Vulnerabilities in Large Language Models and Securing Large Language Models — Bias, Misinformation, and Prompt Attacks.

JailbreakingPrompt attacksMitigation
02

Uncertainty & explainability

From aleatoric to epistemic — a survey of uncertainty quantification techniques in AI — and a comprehensive guide to explainable AI from classical models to LLMs. Also: large language models and cognitive science, the program's most-cited review.

UQXAICognitive science
03

Models in practice — healthcare

LLM benchmarks in medical tasks, a review of clinical trials in drug discovery, and machine-learning prediction of mortality in dialysis patients — the program's clinical anchor in Frontiers in Public Health.

Medical benchmarksClinical trialsDialysis mortality
Latent embedding space
Anchor publication — Use of machine learning models to predict mortality in dialysis patients Frontiers in Public Health 13 · 2025
← The spectrum Traverse again →
B. Peng, Ph.D. — AppCubic · 2026 Imagery : Wellcome Collection CC-BY · RCSB PDB CC0 · computed in-house Google Scholar ↗