Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,49 @@
+# ASCAD V1 Models
+
+Unified repository for all trained models, rank curves, and training metadata from the ASCAD side-channel analysis experiments.
+
+## Overview
+
+This repository contains models trained on the [ASCAD fixed-key dataset](https://github.com/ANSSI-FR/ASCAD) for profiled side-channel attacks targeting all 16 AES key bytes simultaneously.
+
+## Directory Structure
+
+```
+desync{0,50,100}/
+├── mlp/byte{0..15}/          # Single-byte MLP models (200 epochs, batch=100)
+├── cnn/byte{0..15}/          # Single-byte CNN models (150 epochs, batch=200)
+├── lmic_tsbn_v4/             # MTL: LMIC-TSBN baseline (256-class, focal+cosine)
+├── lmic_tsbn_v5e/            # MTL: DTP gamma=2.0
+├── lmic_tsbn_v5f/            # MTL: DTP gamma=5.0
+├── lmic_tsbn_v5g/            # MTL: DTP gamma=5.0 + EMA smoothing
+├── lmic_tsbn_v6/             # MTL: Sigmoid gates + DTP EMA
+└── lmic_tsbn_v7b/            # MTL: Multi-bit (Wu et al.) + DTP (16/16 rank 0!)
+metrics/
+├── experiment_summary.json   # Aggregated results across all experiments
+└── comparison_table.csv      # Quick-reference comparison table
+```
+
+## Per-Model Artifacts
+
+Each model directory contains:
+- `model.h5` — Trained Keras model weights
+- `results.json` — Training configuration, hyperparameters, and rank metrics
+- `rank_curve.npy` (single-byte) or `rank_curve_byte{0..15}.npy` (MTL) — Guessing entropy rank evolution curves
+
+## Experiment Configuration
+
+All experiments use:
+- **Seed:** 42 (fixed for reproducibility and cross-model fairness)
+- **Dataset:** ASCAD fixed-key (50K profiling, 10K attack traces)
+- **Evaluation:** 2000 attack traces, 100 random key shuffles
+- **W&B Project:** `ASCAD_TRAINING_FINAL`
+
+## Key Result
+
+The **LMIC-TSBN V7b** (multi-bit + DTP) model achieves **rank 0 for all 16 AES key bytes** on desync=0, using a single multi-task model with 1M parameters trained for ~40 minutes on an A100 GPU.
+
+## Citation
+
+If you use these models, please cite the ASCAD dataset paper and the Wu et al. (2023) multi-bit approach:
+- Benadjila et al., "Deep learning for side-channel analysis and introduction to ASCAD", IACR ePrint 2018
+- Wu et al., "Breaking the Barrier: Multi-bit Model for Side-Channel Analysis", IACR ePrint 2023/1110