Readability-Robust Code Summarization via Meta Curriculum Learning

TL;DR

This work tackles the challenge of code summarization when source code is poorly readable or obfuscated. It introduces RoFTCodeSum, a fine-tuning framework that combines curriculum learning with model-agnostic meta-learning to train on a curriculum of progressively harder obfuscated code, thereby improving robustness to semantic perturbations while preserving performance on readable code. Empirical results across CodeSearchNet, DeepSeek-Coder, and Qwen2.5-Coder show RoFTCodeSum outperforms baseline fine-tuning and curriculum baselines under obfuscation and even improves performance on original code. The approach offers a generalizable paradigm for robust fine-tuning in code intelligence tasks with practical implications for maintenance, reverse engineering, and security analysis.

Abstract

Code summarization has emerged as a fundamental technique in the field of program comprehension. While code language models have shown significant advancements, the current models and benchmarks are confined to high-readability code, which contains sufficient semantic cues such as function and variable names. In the real world, however, code is often poorly structured or obfuscated, significantly degrading model performance. In this paper, we first empirically evaluate the robustness of state-of-the-art language models on poor-readability code for the task of code summarization, focusing on (1) their effectiveness, (2) the impact of prompt engineering, and (3) the robustness of different variants. Experimental results reveal that state-of-the-art models-including GPT-4o and DeepSeek-V3 experience a substantial performance drop when faced with poorly readable code, and that prompt engineering and reasoning-enhanced models offer limited improvements. Motivated by these findings, we propose RoFTCodeSum, a novel fine-tuning method that enhances the robustness of code summarization against poorly readable code. RoFTCodeSum marries the concepts of curriculum learning and meta-learning: based on the original dataset for fine-tuning, it creates curricular training sets, e.g., obfuscating function names and identifiers from the code, respectively, that have progressive difficulty in code comprehension. In each training step, the approach meta-updates the gradients using these progressively challenging datasets, thereby optimizing both accuracy and readability robustness simultaneously. Experimental results demonstrate that RoFTCodeSum exhibits increased robustness against semantic perturbation while enhancing performance on the original code.

Figures (6)

• Figure 1: A motivating example of how poor readability interferes with code summarization by language models. The obfuscated identifiers are marked in blue.
• Figure 2: Architecture of RoFTCodeSum, illustrating the flow from the original dataset to progressively more challenging obfuscated datasets.
• Figure 3: An example of two curricular datasets. The semantically obfuscated parts are marked in blue.
• Figure 4: Illustration of the MAML algorithm in our method.
• Figure 5: Performance of RoFTCodeSum (DeepSeek-Coder) under different hyperparameters in the semantic-obfuscation curriculum