Language Models (LMs) are widely used in software engineering for code generation, but they may produce code with errors. Rather than repairing the generated code, an alternative way is to address the underlying failures of models. LM repair offers a lightweight solution to this challenge: it requires minimal data, reduces computational costs, and reduces the side effects. Unlike retraining, LM repair focuses on applying tailored updates to targeted neurons, making it ideal for scenarios with limited resources, high-performance demands, or strict safety requirements. In this paper, we propose \ul{S}emantic \ul{T}argeting for \ul{A}nalytical \ul{R}epair (\textsc{STAR}), a pioneering and novel semantic-based optimization approach for repairing LLMs. \textsc{STAR} realizes main operations in LM repair methods in an optimization process, including locating ``buggy neurons'', solving ``neuron patches'', and patching ``buggy neurons''. Correspondingly, it computes the deltas of weight matrix as the prior information to guide optimization; and attributes the targeted layers and neurons leveraging statistical insights. The neuron patches are computed with a solid semantic-based analytical formula, which directly bridges the changes to logits with the deltas of neurons, by steering latent representations. Compared to the prior work of LM repair (\textsc{MINT}) and optimization methods (\textsc{SGD}), \textsc{STAR} integrates their strengths while mitigating their limitations. \textsc{STAR} supports solving multiple failures together, significantly improving the usefulness. Evaluated on three code generation tasks using popular code LMs, \textsc{STAR} demonstrates superior effectiveness. Additionally, \textsc{STAR} exhibits better efficiency. In terms of side effects, namely the balance between generalization and specificity, \textsc{STAR} outperforms prior work by a significant margin.

With the widespread application of LLM-based dialogue systems in daily life, quality assurance has become more important than ever. Recent research has successfully introduced methods to identify unexpected behaviour in single-turn scenarios. However, multi-turn dialogue testing remains underexplored, with the Oracle problem in multi-turn testing posing a persistent challenge for dialogue system developers and researchers. In this paper, we propose MORTAR, a MetamORphic multi-TuRn diAlogue testing appRoach, which mitigates the test oracle problem in the assessment of LLM-based dialogue systems. MORTAR automates the generation of follow-up question-answer (QA) dialogue test cases with multiple dialogue-level perturbations and metamorphic relations. MORTAR employs a novel knowledge graph-based dialogue information model which effectively generates perturbed dialogue test datasets and detects bugs of multi-turn dialogue systems in a low-cost manner. The proposed approach does not require an LLM as a judge, eliminating potential of any biases in the evaluation step. According to the experiment results on multiple LLM-based dialogue systems and comparisons with single-turn metamorphic testing approaches, MORTAR explores more unique bugs in LLM-based dialogue systems, especially for severe bugs that MORTAR detects up to four times more unique bugs than the most effective existing metamorphic testing approach.

Machine learning (ML) for text classification has been widely used in various domains, such as toxicity detection, chatbot consulting, and review analysis. These applications can significantly impact ethics, economics, and human behavior, raising serious concerns about trusting ML decisions. Several studies indicate that traditional metrics, such as model confidence and accuracy, are insufficient to build human trust in ML models. These models often learn spurious correlations during training and predict based on them during inference. In the real world, where such correlations are absent, their performance can deteriorate significantly. To avoid this, a common practice is to test whether predictions are reasonable. Along with this, a challenge known as the trustworthiness oracle problem has been introduced. Due to the lack of automated trustworthiness oracles, the assessment requires manual validation of the decision process disclosed by explanation methods, which is time-consuming and not scalable. We propose TOKI, the first automated trustworthiness oracle generation method for text classifiers, which automatically checks whether the prediction-contributing words are related to the predicted class using explanation methods and word embeddings. To demonstrate its practical usefulness, we introduce a novel adversarial attack method targeting trustworthiness issues identified by TOKI. We compare TOKI with a naive baseline based solely on model confidence using human-created ground truths of 6,000 predictions. We also compare TOKI-guided adversarial attack method with A2T, a SOTA adversarial attack method. Results show that relying on prediction uncertainty cannot distinguish between trustworthy and untrustworthy predictions, TOKI achieves 142% higher accuracy than the naive baseline, and TOKI-guided adversarial attack method is more effective with fewer perturbations than A2T.

Finetuning language models (LMs) is crucial for adapting the models to downstream data and tasks. However, full finetuning is usually costly. Existing work, such as parameter-efficient finetuning (PEFT), often focuses on \textit{how to finetune} but neglects the issue of \textit{where to finetune}. As a pioneering work on answering where to finetune (at the layer level), we conduct a semantic analysis of the LM inference process. We first propose a virtual transition of the latent representation and then trace its factual transition. Based on the deviation in transitions, we estimate the gain of finetuning each model layer, and further, narrow down the scope for finetuning. We perform extensive experiments across well-known LMs and datasets. The results show that our approach is effective and efficient, and outperforms the existing baselines. Our approach is orthogonal to existing efficient techniques, such as PEFT methods, offering practical values on LM finetuning.

Understanding the latent space of language models (LM) is crucial to refining their performance and interpretability. Existing analyses often fall short in providing disentangled (model-centric) insights into LM semantics, and neglect essential aspects of LM adaption. In response, we introduce a pioneering method called vocabulary-defined semantics, which establishes a reference frame within the LM latent space, ensuring disentangled semantic analysis grounded in LM vocabulary. Our approach transcends prior entangled analysis, leveraging LM vocabulary for model-centric insights. Furthermore, we propose a novel technique to compute logits, emphasising differentiability and local isotropy, and introduce a neural clustering module for semantically calibrating data representations during LM adaptation. Through extensive experiments across diverse text understanding datasets, our approach outperforms state-of-the-art methods of retrieval-augmented generation and parameter-efficient finetuning, showcasing its efficacy and broad applicability. Our findings not only shed light on LM mechanics, but also offer practical solutions to enhance LM performance and interpretability.

Large Language Models are successfully adopted in software engineering, especially in code generation. Updating these models with new knowledge is very expensive, and is often required to fully realize their value. In this paper, we propose a novel and effective model editing approach, \textsc{MENT}, to patch LLMs in coding tasks. Based on the mechanism of generative LLMs, \textsc{MENT} enables model editing in next-token predictions, and further supports common coding tasks. \textsc{MENT} is effective, efficient, and reliable. It can correct a neural model by patching 1 or 2 neurons. As the pioneer work on neuron-level model editing of generative models, we formalize the editing process and introduce the involved concepts. Besides, we also introduce new measures to evaluate its generalization ability, and build a benchmark for further study. Our approach is evaluated on three coding tasks, including API-seq recommendation, line-level code generation, and pseudocode-to-code transaction. It outperforms the state-of-the-art by a significant margin on both effectiveness and efficiency measures. In addition, we demonstrate the usages of \textsc{MENT} for LLM reasoning in software engineering. By editing the LLM knowledge with \textsc{MENT}, the directly or indirectly dependent behaviors in the chain-of-thought change accordingly and automatically.

This paper explores the use of a novel form of Hierarchical Graph Neurons (HGN) for in-operation behaviour selection in a swarm of robotic agents. This new HGN is called Robotic-HGN (R-HGN), as it matches robot environment observations to environment labels via fusion of match probabilities from both temporal and intra-swarm collections. This approach is novel for HGN as it addresses robotic observations being pseudo-continuous numbers, rather than categorical values. Additionally, the proposed approach is memory and computation-power conservative and thus is acceptable for use in mobile devices such as single-board computers, which are often used in mobile robotic agents. This R-HGN approach is validated against individual behaviour implementation and random behaviour selection. This contrast is made in two sets of simulated environments: environments designed to challenge the held behaviours of the R-HGN, and randomly generated environments which are more challenging for the robotic swarm than R-HGN training conditions. R-HGN has been found to enable appropriate behaviour selection in both these sets, allowing significant swarm performance in pre-trained and unexpected environment conditions.