paper-with-me

Papers

BatchTopK Sparse Autoencoders

2024-12-09 · Bart Bussmann, Patrick Leask, Neel Nanda

Sparse autoencoders (SAEs) have emerged as a powerful tool for interpreting language model activations by decomposing them into sparse, interpretable features. A popular approach is the TopK SAE, that uses a fixed number of the most active latents per sample to reconstruct the model activations. We introduce BatchTopK SAEs, a training method that improves upon TopK SAEs by relaxing the top-k constraint to the batch-level, allowing for a variable number of latents to be active per sample. As a result, BatchTopK adaptively allocates more or fewer latents depending on the sample, improving reconstruction without sacrificing average sparsity. We show that BatchTopK SAEs consistently outperform TopK SAEs in reconstructing activations from GPT-2 Small and Gemma 2 2B, and achieve comparable performance to state-of-the-art JumpReLU SAEs. However, an advantage of BatchTopK is that the average number of latents can be directly specified, rather than approximately tuned through a costly hyperparameter sweep. We provide code for training and evaluating BatchTopK SAEs at https://github.com/bartbussmann/BatchTopK

📄 PDF Abstract BibTeX arXiv:2412.06410

Code (2)

bartbussmann/batchtopk 공식 구현 pytorch
saprmarks/dictionary_learning pytorch

Tasks

Language ModelingLanguage Modelling

Methods 이 논문이 사용한 방법론

Refunds@Expedia|||How do I get a full refund from Expedia? “How do I get a full refund from Expedia? How do I get a full refund from Expedia? – Call ☎️ +1-(888) 829 (0881) or +1-805-330-4056 or +1-805-330-4056 for Quick Help &…
Attention 설명 없음
Layer Normalization Unlike batch normalization, Layer Normalization directly estimates the normalization statistics from the summed inputs…
Adam 설명 없음
Dropout Dropout is a regularization technique for neural networks that drops a unit (along with connections) at training time with a specified probability $p$ (a common value is…
Attention Dropout Attention Dropout is a type of dropout used in attention-based architectures, where elements are randomly dropped out of the…
Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Dense Connections Dense Connections, or Fully Connected Connections, are a type of layer in a deep neural network that use a linear operation where every input is connected to every output…

Similar Papers 제목 키워드 기반

Distribution-Aware Feature Selection for SAEs

2025-08-29 · Narmeen Oozeer, Nirmalendu Prakash, Michael Lan, Alice Rigg 외 arxiv

Sparse autoencoders (SAEs) decompose neural activations into interpretable features. A widely adopted variant, the TopK SAE, reconstructs each token from its K most active latents. However, this approach is inefficient, …

Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders

2026-06-08 · Nikita Koriagin, Georgii Aparin, Nikita Balagansky, Daniil Gavrilov arxiv

Language models increasingly serve as the backbone of text-to-speech (TTS) systems, yet we understand little about the representations they build when text and generated speech tokens share a single residual stream. We t…

Sparse Crosscoders for diffing MoEs and Dense models

2026-03-06 · Marmik Chaudhari, Nishkal Hundia, Idhant Gulati arxiv

Mixture of Experts (MoE) achieve parameter-efficient scaling through sparse expert routing, yet their internal representations remain poorly understood compared to dense models. We present a systematic comparison of MoE …

Sparse but not Simpler: A Multi-Level Interpretability Analysis of Vision Transformers

2026-03-16 · Siyu Zhang arxiv

Sparse neural networks are often hypothesized to be more interpretable than dense models, motivated by findings that weight sparsity can produce compact circuits in language models. However, it remains unclear whether st…

Decompose Sparsely Where You Should, Absorb Densely Where You Should No

2026-06-12 · Ruixuan Deng, Zehao Jin, Zekun Wang, Zihan Dong arxiv

Sparse autoencoders (SAEs) are typically trained to reconstruct the \textbf{entire} residual stream through a sparse dictionary, implicitly assuming that all activation content is amenable to sparse, monosemantic decompo…