# results_clf_noise_aug_ft

Fine-tune published **SW2** `vlaai_ws` 10 s classifiers from held-*/all5 checkpoints with train-only **EEG noise / channel-drop / SpecAugment-style** curriculum (noise=0.1, p_chan=0.2, p_cluster=0.3, p_tilt=0.0, p_time_mask=0.0, p_freq_mask=0.0). `p_switch=0.5`, `clf_p_switch2=0.5` (= published sw2; **not** denser switch). Architecture unchanged. **Not** hardswitch/shorthold/switch-dur, **not** MixStyle, **not** tip Acc PP / emission-shift.

Prep `eaP_n5_t60`, FT lr=0.00025, 4×250 steps, seeds=[0]. HMM p_stay 0.994 scale 0.8.

Code: `clf_noise_aug_ft_lodo.py`. JSON: `/root/disk_3tb/eeg-thao/results/research/cfnull_l0p1.json`. Ckpts: `models/clf_noise_aug_ft/s*/`.

## Headline Acc

Selector stream: **seed0**

| Set | H / Acc | vs tip 53.53 / Carta hp30 60.54 |
|---|---|---|
| AASD | 0.6367 / **52.35** | ΔAcc -1.18 |
| Carta2026 | 0.6428 / **58.57** | ΔAcc -1.97 |
| LODO mean | 0.7880 / 71.74 | |

AASD Acc>60? **NO**

## Per-seed LODO + external

| Seed | LODO H/Acc | AASD H/Acc | Carta H/Acc |
|---|---|---|---|
| 0 | 0.7880/71.74 | 0.6367/52.35 | 0.6428/58.57 |

## Targets

| Target | Result |
|---|---|
| AASD Acc vs tip 53.53 | Acc **52.35** (Δ-1.18) |
| AASD Acc>60? | NO |
| Carta vs linear_hp30 60.54 | Acc **58.57** |
| vs linear_hp30 AASD 53.12 | Δ-0.77 |

## Verdict

EEG noise/SpecAugment FT from published SW2: AASD Acc 52.35 vs tip 53.53 (Δ-1.18) via seed0.

## NEXT IDEA

If Acc flat/~53: stronger EEG noise/channel-drop/SpecAug FT alone does not close the real-switch Acc gap (same envelope as switch-schedule FT). Prefer **speech-xattn-dec**: multi-head cross-attn decoder with speech-envelope tokens as KV and EEG latents as Q (end-to-end Acc LODO) — ≠ temp-attn-spk (self-attn on EEG+speech concat), ≠ speech-resid (FiLM/additive residual), ≠ aux-env-spk (MSE recon aux). Not emis-shift / tip gate / SSL / Path A / SpecViT.

## AVGC note (unused; do not burn GPU)

AVGC tip gap ≈−0.027 H: unused idea for coordinator — causal *per-dataset* temperature on frozen focal↔synth fuse (train-only T for AVGC fold only; ≠ Mixup/focal/DANN/SSL).
