Large Language Models are Interpretable Learners
Ruochen Wang, Si Si, Felix Yu, Dorothea Wiesmann, Cho-Jui, Hsieh, Inderjit Dhillon
TL;DR
This paper introduces LLM-based Symbolic Programs (LSPs), a method combining large language models and symbolic rules to create interpretable, accurate, and transferable decision-making models that outperform traditional approaches.
Contribution
The paper proposes a novel LLM-based symbolic programming framework with a divide-and-conquer training approach, enhancing interpretability and performance in diverse tasks.
Findings
LSPs outperform traditional neurosymbolic methods.
LSPs provide interpretable and transferable knowledge.
LSPs generalize well to out-of-distribution data.
Abstract
The trade-off between expressiveness and interpretability remains a core challenge when building human-centric predictive models for classification and decision-making. While symbolic rules offer interpretability, they often lack expressiveness, whereas neural networks excel in performance but are known for being black boxes. In this paper, we show a combination of Large Language Models (LLMs) and symbolic programs can bridge this gap. In the proposed LLM-based Symbolic Programs (LSPs), the pretrained LLM with natural language prompts provides a massive set of interpretable modules that can transform raw input into natural language concepts. Symbolic programs then integrate these modules into an interpretable decision rule. To train LSPs, we develop a divide-and-conquer approach to incrementally build the program from scratch, where the learning process of each step is guided by LLMs.…
Peer Reviews
Decision·ICLR 2025 Poster
The main strengths of the paper are : + **Clarity** : the paper is well written with a clear and well motivated idea. The paper provides nice illustration that enable a clear understanding of the proposed approach. The addressed research question are also very clear. Moreover, a thorough description of the proposed framework, including ablation studies and code for reproducibility is provided. + **Originality** : the main originality of the paper is to use prompting and the ability of summariza
While the core idea is interesting, the paper has several limitations: + a first small concerns is about the title of the paper which is for me not really aligned on the content. The authors use LLMs as learnable building blocks for NSPs but are not interpretable as their own. The title of the paper should be slightly changed to better reflect the true claims of the paper. + **lack of clear definition of the notion of interpretability and of the underlying assumptions** : as the authors tackle t
1. A new benchmark for interpretable learning tasks, for both vision and text scenarios. 2. Design a LLM-symbolic program generation algorithm (called LSP by authors) by a program structure search procedure. 3. Evaluation shows that the LSP significantly outperforms the baselines in the authors' evaluation settings.
1. As the authors say, LLM's responsibility is to make decisions in each LLM module in their llm-symbolic programs. Thus, generating a correct LSP relies on the LLM's ability. I am curious whether the proposed method could correspond with other open-source LLMs that could be deployed locally.
The approach to integrating LLMs into DSL-based neuro-symbolic programming approaches is interesting, although the idea itself may be straightforward. By combining Large Language Models (LLMs) with symbolic programs, the authors effectively leverage the strengths of both methods: LLMs for creating interpretable natural language concepts and symbolic programs for incorporating these into decision rules. This hybrid approach not only enhances model interpretability without sacrificing performance
I found the manuscript somewhat challenging to follow due to the absence of consistent examples within the main text. Currently, major examples are relegated to the supplementary materials. To enhance clarity, I recommend incorporating one or two examples directly into the main manuscript and referencing them throughout. Moreover, a clear problem statement would significantly improve comprehension. Specifically, outlining the input and output parameters before discussing the methodology would b
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsTopic Modeling · Natural Language Processing Techniques
MethodsSparse Evolutionary Training