min-entropy

The pytorch implementation of the Min-Entropy Latent Model for Weakly Supervised Object Detection

Min-Entropy Latent Model for Weakly Supervised Object Detection

Code repository for "Machine Learning Predictors for Min-Entropy Estimation" (arXiv:2406.19983). Implements RCNN and GPT-2 models for entropy prediction in RNGs, with data generation, training pipelines, and analysis scripts.

Data and code to design and evaluate the PLL-based true random number generator according to the paper "Enhancing Quality and Security of the PLL-TRNG" (published and awarded at TCHES 2023).

Field guide to randomness, entropy and RNGs: PRNG vs TRNG vs CSPRNG, the NIST SP 800-90A/B/C pipeline, min-entropy, conditioning, real-world seeding failures, and why Cloudflare's lava lamps harvest entropy instead of creating it. Born from a collaborative cross-review of a YouTube thread.

Formal verification of NIST SP 800-90B entropy estimators in Lean 4. Companion project to lopt's TRNG service; theorems intended for upstream contribution to mathlib.

Quarto strumento della stessa serie (EntropyPipeline / entropy-extractor / entropy-extractor-raw-2photo), con uno scopo diverso dagli altri tre: non estrae entropia, la cerca di smentire. Verifica l'eventuale presenza di pattern anche se le sorgenti arrivano da fonti diverse.

Strumento didattico standalone, offline, senza dipendenze esterne per sperimentare un pipeline completo di entropy conditioning: raccolta di bit grezzi → stima di min-entropia (in stile NIST SP 800-90B) → estrazione del bias (Peres/von Neumann Extractor) → hashing universale (Toeplitz, GF(2), con bound del Leftover Hash Lemma) → output finale

Strumento standalone, a singolo file HTML per estrarre entropia crittografica da una sorgente grezza unica (dump audio RAW/PCM, video, dump di un TRNG hardware, csv), analizzandola in 4 finestre (Primaria / Mediana / Terminale / Spettro Totale) e selezionando automaticamente quella migliore e statisticamente indipendente.

Estrattore di entropia standalone che ricava bit casuali dal rumore elettronico di un sensore fotografico, eliminando i dark frame da due scatti identici identici

Browser-based QRNG demo — NIST SP 800-90B. Min-entropy is what an optimal guesser faces; Shannon is the marketing number. Slide the detector bias and watch them diverge. Von Neumann fixes bias but not correlation — feed it a correlated stream and watch it output garbage while claiming success. Real Toeplitz extraction over GF(2). No backends.