자기생성 피드백이 테스트 시간 학습을 불안정화: 장기 적응의 인과 분해
Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation
- 매체
- Hugging Face
- 게시
- 중요도
- 3/5
- 화제 지수
- 10
한국어 요약
테스트 시간 학습(TTT)에서 모델이 자신의 출력으로부터 학습할 때, 가중치 업데이트가 다음 학습 예제를 생성하는 모델을 변경하면서 불안정성이 발생한다는 연구 결과다. 128K 토큰 스트림에서 생성된 텍스트 업데이트가 독립적인 인간 작성 텍스트의 예측 성능을 악화시키는 현상을 확인했으며, 고정 생성(Fixed Generation)과 정착(Settlement) 기법으로 이를 완화할 수 있음을 보였다.
원문 발췌 · English
Test-time training (TTT) lets a model store information in its weights during inference. When the model learns from its own output, however, each update also changes the model that generates the next training example. Across 128K-token streams, retaining generated-text updates worsens prediction on independent human-written text with three TTT-E2E model configurations (labeled 125M, 760M, and 3B). The same failure occurs when Adam updates Qwen3-4B's existing weights. The same update mechanisms can improve on real text, so writing itself is not the failure. Three matched comparisons trace the causal pathway. Fixed Generation removes over 98% of the damage at 125M and 760M by using a frozen model to generate training chunks. Recorded Replay separates the loss caused by reading degraded text from the additional loss stored by updating on it. A paired one-update comparison then shows the local conflict: an update predicts its source better but new real text worse. This cost grows after Closed Loop adaptation, with a few trajectories accounting for most large failures. Finally, Settlement evaluates the candidate state on independent real text before commitment. It leaves mean endpoint gaps of 0.07 and -0.02 nats at 125M and 760M while retaining real-text adaptation. These results motivate checking prediction on independent evidence before retaining an update.