Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla

• URL: http://arxiv.org/abs/2307.09458v3
• Date: Mon, 24 Jul 2023 08:32:40 GMT
• Title: Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla
• Authors: Tom Lieberum, Matthew Rahtz, J\'anos Kram\'ar, Neel Nanda, Geoffrey Irving, Rohin Shah, Vladimir Mikulik
• Abstract summary: We present a case study of circuit analysis in the 70B Chinchilla model. We investigate Chinchilla's capability to identify the correct answer emphlabel given knowledge of the correct answer emphtext We study the correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results.
• Score: 6.625597238953314
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: \emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \emph{label} given knowledge of the correct answer \emph{text}. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.

Related papers

• Answer, Assemble, Ace: Understanding How Transformers Answer Multiple Choice Questions [103.20281438405111]
  Multiple-choice question answering (MCQA) is a key competence of performant transformer language models. We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information. We show that prediction of a specific answer symbol is causally attributed to a single middle layer, and specifically its multi-head self-attention mechanism.
  arXiv  Detail & Related papers  (2024-07-21T00:10:23Z)
• Open-Set Knowledge-Based Visual Question Answering with Inference Paths [79.55742631375063]
  The purpose of Knowledge-Based Visual Question Answering (KB-VQA) is to provide a correct answer to the question with the aid of external knowledge bases. We propose a new retriever-ranker paradigm of KB-VQA, Graph pATH rankER (GATHER for brevity) Specifically, it contains graph constructing, pruning, and path-level ranking, which not only retrieves accurate answers but also provides inference paths that explain the reasoning process.
  arXiv  Detail & Related papers  (2023-10-12T09:12:50Z)
• Locate before Answering: Answer Guided Question Localization for Video Question Answering [70.38700123685143]
  LocAns integrates a question locator and an answer predictor into an end-to-end model. It achieves state-of-the-art performance on two modern long-term VideoQA datasets.
  arXiv  Detail & Related papers  (2022-10-05T08:19:16Z)
• Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering [124.16250115608604]
  We present Science Question Answering (SQA), a new benchmark that consists of 21k multimodal multiple choice questions with a diverse set of science topics and annotations of their answers with corresponding lectures and explanations. We show that SQA improves the question answering performance by 1.20% in few-shot GPT-3 and 3.99% in fine-tuned UnifiedQA. Our analysis further shows that language models, similar to humans, benefit from explanations to learn from fewer data and achieve the same performance with just 40% of the data.
  arXiv  Detail & Related papers  (2022-09-20T07:04:24Z)
• ListReader: Extracting List-form Answers for Opinion Questions [18.50111430378249]
  ListReader is a neural ex-tractive QA model for list-form answer. In addition to learning the alignment between the question and content, we introduce a heterogeneous graph neural network. Our model adopts a co-extraction setting that can extract either span- or sentence-level answers.
  arXiv  Detail & Related papers  (2021-10-22T10:33:08Z)
• A Semantic-based Method for Unsupervised Commonsense Question Answering [40.18557352036813]
  Unsupervised commonsense question answering is appealing since it does not rely on any labeled task data. We present a novel SEmantic-based Question Answering method (SEQA) for unsupervised commonsense question answering.
  arXiv  Detail & Related papers  (2021-05-31T08:21:52Z)
• Select, Substitute, Search: A New Benchmark for Knowledge-Augmented Visual Question Answering [35.855792706139525]
  Multimodal IR, spanning text corpus, knowledge graph and images, is of much recent interest. A surprisingly large fraction of queries do not assess the ability to integrate cross-modal information. We build a new data set and challenge around a key structural idiom in OKVQA,viz., S3.
  arXiv  Detail & Related papers  (2021-03-09T17:19:50Z)
• A Wrong Answer or a Wrong Question? An Intricate Relationship between Question Reformulation and Answer Selection in Conversational Question Answering [15.355557454305776]
  We show that question rewriting (QR) of the conversational context allows to shed more light on this phenomenon. We present the results of this analysis on the TREC CAsT and QuAC (CANARD) datasets.
  arXiv  Detail & Related papers  (2020-10-13T06:29:51Z)
• Counterfactual Variable Control for Robust and Interpretable Question Answering [57.25261576239862]
  Deep neural network based question answering (QA) models are neither robust nor explainable in many cases. In this paper, we inspect such spurious "capability" of QA models using causal inference. We propose a novel approach called Counterfactual Variable Control (CVC) that explicitly mitigates any shortcut correlation.
  arXiv  Detail & Related papers  (2020-10-12T10:09:05Z)
• Robust Question Answering Through Sub-part Alignment [53.94003466761305]
  We model question answering as an alignment problem. We train our model on SQuAD v1.1 and test it on several adversarial and out-of-domain datasets.
  arXiv  Detail & Related papers  (2020-04-30T09:10:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.