![[2602.16829] Learning under noisy supervision is governed by a feedback-truth gap](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16829] Learning under noisy supervision is governed by a feedback-truth gap
Summary
This paper explores how learning under noisy supervision is influenced by a feedback-truth gap, demonstrating its effects across various neural networks and human learning scenarios.
Why It Matters
Understanding the feedback-truth gap is crucial for improving machine learning models and human learning processes, especially in environments with noisy data. This research provides insights that can enhance model training and educational strategies.
Key Takeaways
- The feedback-truth gap occurs when feedback is processed faster than task evaluation.
- Different systems regulate this gap in unique ways, affecting learning outcomes.
- Neural networks and human learners exhibit distinct behaviors in response to noisy supervision.
Computer Science > Machine Learning arXiv:2602.16829 (cs) [Submitted on 18 Feb 2026] Title:Learning under noisy supervision is governed by a feedback-truth gap Authors:Elan Schonfeld, Elias Wisnia View a PDF of the paper titled Learning under noisy supervision is governed by a feedback-truth gap, by Elan Schonfeld and Elias Wisnia View PDF HTML (experimental) Abstract:When feedback is absorbed faster than task structure can be evaluated, the learner will favor feedback over truth. A two-timescale model shows this feedback-truth gap is inevitable whenever the two rates differ and vanishes only when they match. We test this prediction across neural networks trained with noisy labels (30 datasets, 2,700 runs), human probabilistic reversal learning (N = 292), and human reward/punishment learning with concurrent EEG (N = 25). In each system, truth is defined operationally: held-out labels, the objectively correct option, or the participant's pre-feedback expectation - the only non-circular reference decodable from post-feedback EEG. The gap appeared universally but was regulated differently: dense networks accumulated it as memorization; sparse-residual scaffolding suppressed it; humans generated transient over-commitment that was actively recovered. Neural over-commitment (~0.04-0.10) was amplified tenfold into behavioral commitment (d = 3.3-3.9). The gap is a fundamental constraint on learning under noisy supervision; its consequences depend on the regulation each system empl...