Large Language Models (LLMs) have recently revolutionized the NLP field, while they still fall short in some specific down-stream tasks. In the work, we focus on utilizing LLMs to perform machine translation, where we observe that two patterns of errors frequently occur and drastically affect the translation quality: language mismatch and repetition. The work sets out to explore the potential for mitigating these two issues by leveraging model editing methods, e.g., by locating Feed-Forward Network (FFN) neurons or something that are responsible for the errors and deactivating them in the inference time. We find that directly applying such methods either limited effect on the targeted errors or has significant negative side-effect on the general translation quality, indicating that the located components may also be crucial for ensuring machine translation with LLMs on the rails. To this end, we propose to refine the located components by fetching the intersection of the locating results under different language settings, filtering out the aforementioned information that is irrelevant to targeted errors. The experiment results empirically demonstrate that our methods can effectively reduce the language mismatch and repetition ratios and meanwhile enhance or keep the general translation quality in most cases.

QLoRA reduces the memory-cost of fine-tuning a large language model (LLM) with LoRA by quantizing the base LLM. However, quantization introduces quantization errors that negatively impact model performance after fine-tuning. In this paper we introduce QuAILoRA, a quantization-aware initialization for LoRA that mitigates this negative impact by decreasing quantization errors at initialization. Our method spends a small amount of computational overhead to compute this quantization-aware initialization, without increasing the memory-cost of fine-tuning. We evaluate our method on several causal language modeling and downstream evaluation tasks using several different model sizes and families. We observe that almost all LLMs fined-tuned with QuAILoRA achieve better validation perplexity. When evaluated on downstream tasks, we find that QuAILoRA yields improvements proportional to the negative effect of quantization error.

Paleontology, the study of past life, fundamentally relies on fossils to reconstruct ancient ecosystems and understand evolutionary dynamics. Trilobites, as an important group of extinct marine arthropods, offer valuable insights into Paleozoic environments through their well-preserved fossil records. Reconstructing trilobite behaviour from static fossils will set new standards for dynamic reconstructions in scientific research and education. Despite the potential, current computational methods for this purpose like text-to-video (T2V) face significant challenges, such as maintaining visual realism and consistency, which hinder their application in science contexts. To overcome these obstacles, we introduce an automatic T2V prompt learning method. Within this framework, prompts for a fine-tuned video generation model are generated by a large language model, which is trained using rewards that quantify the visual realism and smoothness of the generated video. The fine-tuning of the video generation model, along with the reward calculations make use of a collected dataset of 9,088 Eoredlichia intermedia fossil images, which provides a common representative of visual details of all class of trilobites. Qualitative and quantitative experiments show that our method can generate trilobite videos with significantly higher visual realism compared to powerful baselines, promising to boost both scientific understanding and public engagement.

Large Language Models (LLMs) and Code-LLMs (CLLMs) have significantly improved code generation, but, they frequently face difficulties when dealing with challenging and complex problems. However, retrieval models often fail to find most relevant context, and generation models, with limited context capacity, can hallucinate when given irrelevant data. We present a novel framework that leverages a Programming Knowledge Graph (PKG) to semantically represent and retrieve code. This approach enables fine-grained code retrieval by focusing on the most relevant segments while reducing irrelevant context through a tree-pruning technique. PKG is coupled with a re-ranking mechanism to reduce even more hallucinations by selectively integrating non-RAG solutions. We propose two retrieval approaches--block-wise and function-wise--based on the PKG, optimizing context granularity. Evaluations on the HumanEval and MBPP benchmarks show our method improves pass@1 accuracy by up to 20%, and outperforms state-of-the-art models by up to 34% on MBPP. Our contributions include PKG-based retrieval, tree pruning to enhance retrieval precision, a re-ranking method for robust solution selection and a Fill-in-the-Middle (FIM) enhancer module for automatic code augmentation with relevant comments and docstrings. Large Language Models (LLMs) have significantly improved the performance of tasks related to code, such as code generation (Huang et al., 2023; Roziere et al., 2023a; Li et al., 2023; Wang et al., 2023). As code-related models continue to emerge rapidly (Chen et al., 2021; Li et al., 2023; 2022; Roziere et al., 2023a; Zhu et al., 2024), most of these models rely on a natural language-to-code (NL-to-Code) paradigm, which often lacks the ability to leverage existing contextual information (Wang et al., 2024). Generating a solution from scratch, without access to supplementary context, poses significant challenges (Wang et al., 2024), even for humans (Zhong et al., 2024).

Sonar radar captures visual representations of underwater objects and structures using sound wave reflections, making it essential for exploration, mapping, and continuous surveillance in wild ecosystems. Real-time analysis of sonar data is crucial for time-sensitive applications, including environmental anomaly detection and in-season fishery management, where rapid decision-making is needed. However, the lack of both relevant datasets and pre-trained DNN models, coupled with resource limitations in wild environments, hinders the effective deployment and continuous operation of live sonar analytics. We present SALINA, a sustainable live sonar analytics system designed to address these challenges. SALINA enables real-time processing of acoustic sonar data with spatial and temporal adaptations, and features energy-efficient operation through a robust energy management module. Deployed for six months at two inland rivers in British Columbia, Canada, SALINA provided continuous 24/7 underwater monitoring, supporting fishery stewardship and wildlife restoration efforts. Through extensive real-world testing, SALINA demonstrated an up to 9.5% improvement in average precision and a 10.1% increase in tracking metrics. The energy management module successfully handled extreme weather, preventing outages and reducing contingency costs. These results offer valuable insights for long-term deployment of acoustic data systems in the wild.

Regular documentation of progress notes is one of the main contributors to clinician burden. The abundance of structured chart information in medical records further exacerbates the burden, however, it also presents an opportunity to automate the generation of progress notes. In this paper, we propose a task to automate progress note generation using structured or tabular information present in electronic health records. To this end, we present a novel framework and a large dataset, ChartPNG, for the task which contains $7089$ annotation instances (each having a pair of progress notes and interim structured chart data) across $1616$ patients. We establish baselines on the dataset using large language models from general and biomedical domains. We perform both automated (where the best performing Biomistral model achieved a BERTScore F1 of $80.53$ and MEDCON score of $19.61$) and manual (where we found that the model was able to leverage relevant structured data with $76.9\%$ accuracy) analyses to identify the challenges with the proposed task and opportunities for future research.

In this paper, we conduct a comprehensive SWOT analysis of prompt engineering techniques within the realm of Large Language Models (LLMs). Emphasizing linguistic principles, we examine various techniques to identify their strengths, weaknesses, opportunities, and threats. Our findings provide insights into enhancing AI interactions and improving language model comprehension of human prompts. The analysis covers techniques including template-based approaches and fine-tuning, addressing the problems and challenges associated with each. The conclusion offers future research directions aimed at advancing the effectiveness of prompt engineering in optimizing human-machine communication.

In the era of large language models, parameter-efficient fine-tuning (PEFT) has been extensively studied. However, these approaches usually rely on the space domain, which encounters storage challenges especially when handling extensive adaptations or larger models. The frequency domain, in contrast, is more effective in compressing trainable parameters while maintaining the expressive capability. In this paper, we propose a novel Selective Discrete Cosine Transformation (sDCTFT) fine-tuning scheme to push this frontier. Its general idea is to exploit the superior energy compaction and decorrelation properties of DCT to improve both model efficiency and accuracy. Specifically, it projects the weight change from the low-rank adaptation into the discrete cosine space. Then, the weight change is partitioned over different levels of the discrete cosine spectrum, and the most critical frequency components in each partition are selected. Extensive experiments on four benchmark datasets demonstrate the superior accuracy, reduced computational cost, and lower storage requirements of the proposed method over the prior arts. For instance, when performing instruction tuning on the LLaMA3.1-8B model, sDCTFT outperforms LoRA with just 0.05M trainable parameters compared to LoRA's 38.2M, and surpasses FourierFT with 30\% less trainable parameters. The source code will be publicly available.

In the era of large language models (LLMs), detecting pretraining data has been increasingly important due to concerns about fair evaluation and ethical risks. Current methods differentiate members and non-members by designing scoring functions, like Perplexity and Min-k%. In this paper, we first explore the benefits of unseen data, which can be easily collected after the release of the LLM. We find that the perplexities of LLMs perform differently for members and non-members, after fine-tuning with a small amount of previously unseen data. In light of this, we introduce a novel and effective method termed Fine-tuned Score Deviation (FSD), which improves the performance of current scoring functions for pretraining data detection. In particular, we propose to measure the deviation distance of current scores after fine-tuning on a small amount of unseen data within the same domain. In effect, using a few unseen data can largely decrease the scores of all non-members, leading to a larger deviation distance than members. Extensive experiments demonstrate the effectiveness of our method, significantly improving the AUC score on common benchmark datasets across various models. The impressive performance of large language models (LLMs) arises from large-scale pretraining on massive datasets collected from the internet (Achiam et al., 2023; Touvron et al., 2023b). But, model developers are often reluctant to disclose detailed information about the pretraining datasets, raising significant concerns regarding fair evaluation and ethical risks. Specifically, Recent studies reveal that the pretraining corpus may inadvertently include data from evaluation benchmarks (Sainz et al., 2023; Balloccu et al., 2024), making it difficult to assess the practical capability of LLMs. Considering the vast size of the pretraining dataset and the single iteration of pretraining, it has been increasingly important and challenging to detect pretraining data, which determines whether a piece of text is part of the pretraining dataset.

Logical errors in programming don't raise compiler alerts, making them hard to detect. These silent errors can disrupt a program's function or cause run-time issues. Their correction requires deep insight into the program's logic, highlighting the importance of automated detection and repair. In this paper, we introduce LecPrompt to localize and repair logical errors, an prompt-based approach that harnesses the capabilities of CodeBERT, a transformer-based large language model trained on code. First, LecPrompt leverages a large language model to calculate perplexity and log probability metrics, pinpointing logical errors at both token and line levels. Through statistical analysis, it identifies tokens and lines that deviate significantly from the expected patterns recognized by large language models, marking them as potential error sources. Second, by framing the logical error correction challenge as a Masked Language Modeling (MLM) task, LecPrompt employs CodeBERT to autoregressively repair the identified error tokens. Finally, the soft-prompt method provides a novel solution in low-cost scenarios, ensuring that the model can be fine-tuned to the specific nuances of the logical error correction task without incurring high computational costs. To evaluate LecPrompt's performance, we created a method to introduce logical errors into correct code and applying this on QuixBugs to produce the QuixBugs-LE dataset. Our evaluations on the QuixBugs-LE dataset for both Python and Java highlight the impressive capabilities of our method, LecPrompt. For Python, LecPrompt achieves a noteworthy 74.58% top-1 token-level repair accuracy and 27.4% program-level repair accuracy. In Java, LecPrompt delivers a 69.23\% top-1 token-level repair accuracy and 24.7% full program-level repair accuracy.