paper-with-me

Papers

Backward Lens: Projecting Language Model Gradients into the Vocabulary Space

2024-02-20 · Shahar Katz, Yonatan Belinkov, Mor Geva, Lior Wolf

Understanding how Transformer-based Language Models (LMs) learn and recall information is a key goal of the deep learning community. Recent interpretability methods project weights and hidden states obtained from the forward pass to the models' vocabularies, helping to uncover how information flows within LMs. In this work, we extend this methodology to LMs' backward pass and gradients. We first prove that a gradient matrix can be cast as a low-rank linear combination of its forward and backward passes' inputs. We then develop methods to project these gradients into vocabulary items and explore the mechanics of how new information is stored in the LMs' neurons.

📄 PDF Abstract BibTeX arXiv:2402.12865

Code (0)

등록된 구현이 없습니다.

Tasks

Language ModelingLanguage Modelling

Similar Papers 제목 키워드 기반

Decoding Vision Transformers: the Diffusion Steering Lens

2025-04-18 · Ryota Takatsuki, Sonia Joseph, Ippei Fujisawa, Ryota Kanai

Logit Lens is a widely adopted method for mechanistic interpretability of transformer-based language models, enabling the analysis of how internal representations evolve across layers by projecting them into the output v…

IG-Lens: Exact Additive Probability Attribution Across Transformer Layers via Telescoping Integrated Gradients

2026-06-29 · Duc Anh Nguyen arxiv

We ask a simple question about decoder-only transformers: between which two layers is the probability of a predicted token actually produced? Existing layer-wise readout tools answer only approximately. The logit lens an…

DistillLens: Symmetric Knowledge Distillation Through Logit Lens

2026-02-14 · Manish Dhakal, Uthman Jinadu, Anjila Budathoki, Rajshekhar Sunderraman 외 arxiv

Standard Knowledge Distillation (KD) compresses Large Language Models (LLMs) by optimizing final outputs, yet it typically treats the teacher's intermediate layer's thought process as a black box. While feature-based dis…

Knowledge Distillation

Altering Backward Pass Gradients improves Convergence

2021-11-24 · Bishshoy Das, Milton Mondal, Brejesh lall, Shiv Dutt Joshi 외

In standard neural network training, the gradients in the backward pass are determined by the forward pass. As a result, the two stages are coupled. This is how most neural networks are trained currently. However, gradie…

Generative Model Inversion Through the Lens of the Manifold Hypothesis

2025-09-24 · Xiong Peng, Bo Han, Fengfei Yu, Tongliang Liu 외 arxiv

Model inversion attacks (MIAs) aim to reconstruct class-representative samples from trained models. Recent generative MIAs utilize generative adversarial networks to learn image priors that guide the inversion process, y…