Jet Expansions of Residual Computation
We introduce a framework for expanding residual computational graphs using jets, operators that generalize truncated Taylor series. Our method provides a systematic approach to disentangle contributions of different computational paths to model predictions. In contrast to existing techniques such as distillation, probing, or early decoding, our expansions rely solely on the model itself and requires no data, training, or sampling from the model. We demonstrate how our framework grounds and subsumes logit lens, reveals a (super-)exponential path structure in the recursive residual depth and opens up several applications. These include sketching a transformer large language model with $n$-gram statistics extracted from its computations, and indexing the models' levels of toxicity knowledge. Our approach enables data-free analysis of residual computation for model interpretability, development, and evaluation.
Code (0)
등록된 구현이 없습니다.
Tasks
Language ModelingLanguage ModellingLarge Language ModelSimilar Papers 제목 키워드 기반
Stochastic spectral embedding
Constructing approximations that can accurately mimic the behavior of complex models at reduced computational costs is an important aspect of uncertainty quantification. Despite their flexibility and efficiency, classica…
Uncertainty QuantificationEnhanced Residual Networks for Context-based Image Outpainting
Although humans perform well at predicting what exists beyond the boundaries of an image, deep models struggle to understand context and extrapolation through retained information. This task is known as image outpainting…
Image OutpaintingHierarchical Bipartite Graph Convolution Networks
Recently, graph neural networks have been adopted in a wide variety of applications ranging from relational representations to modeling irregular data domains such as point clouds and social graphs. However, the space of…
Graph Neural NetworkFunctional Expansions
Path dependence is omnipresent in many disciplines such as engineering, system theory and finance. It reflects the influence of the past on the future, often expressed through functionals. However, non-Markovian problems…
Adaptive Basis Function Selection for Computationally Efficient Predictions
Basis Function (BF) expansions are a cornerstone of any engineer's toolbox for computational function approximation which shares connections with both neural networks and Gaussian processes. Even though BF expansions are…
Gaussian Processes