Predictive Closure:JEPA 特徵選擇的集合層面刻畫
Predictive Closure: A Set-Level Characterization of JEPA's Feature Selection
研究提出 predictive closure,指出 JEPA 保留哪些特徵取決於特徵集合之間的預測關係,而非個別特徵是否容易預測。受控實驗顯示,難以預測但有助預測其他特徵的特徵,可以得到更好的表徵;改變一項特徵的預測關係,也會影響其他特徵的表徵。透過輔助目標或向 predictor 提供側資訊,亦可間接引導特徵選擇。
Published on Oct 19
Authors:
,
Abstract
JEPA feature selection depends on predictive relationships between features rather than individual predictability, allowing unpredictable but relationally useful features to be retained and enabling indirect steering of representations.
In joint-embedding predictive architectures (JEPA), an encoder and a predictor are trained together to produce a representation that minimizes the error in predicting the representation of the next observation. Because the representations of both the current and the next observation are learned, it is unclear which features they will include. Earlier characterizations describe the features JEPA selects in terms of their individual predictability, raising the concern that unpredictable features, like a traffic light, will not be encoded into the representation. However, the prediction error is also affected by the relationships between features, e.g., the light's color helps predict whether the traffic will stop. We propose the predictive closure characterization, which accounts for these relationships. Based on a reformulation of the JEPA training objective, we show that selection depends on properties of feature sets. This view agrees with earlier works when features evolve independently, but makes new predictions when they do not, for instance that a feature like the traffic light can still be retained. We test this characterization in controlled experiments, which show that unpredictable features can become better represented when they help predict other features, and that changing one feature's predictive relationships can affect how well others are represented. We also demonstrate implications for steering feature selection: targeting one feature with an auxiliary objective, or supplying it to the predictor as side information, can strengthen the representation of another feature whose future values it helps predict, so steering can be indirect.
View arXiv page View PDF GitHub Add to collection
Get this paper in your agent:
hf papers read 2610.27364
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash
Models citing this paper 0
No model linking this paper
Cite arxiv.org/abs/2610.27364 in a model README.md to link it from this page.
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2610.27364 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2610.27364 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co