Understanding Layer Significance in LLM Alignment

Aligning large language models (LLMs) through fine-tuning is essential for tailoring them to specific applications. Therefore, understanding what LLMs learn during the alignment process is crucial. Recent studies suggest that alignment primarily adjusts a model's presentation style rather than its foundational knowledge, indicating that only certain components of the model are significantly impacted. To delve deeper into LLM alignment, we propose to identify which layers within LLMs are most critical to the alignment process, thereby uncovering how alignment influences model behavior at a granular level. We propose a novel approach to identify the important layers for LLM alignment (ILA). It involves learning a binary mask for each incremental weight matrix in the LoRA algorithm, indicating the significance of each layer. ILA consistently identifies important layers across various alignment datasets, with nearly 90% overlap even with substantial dataset differences, highlighting fundamental patterns in LLM alignment. Experimental results indicate that freezing non-essential layers improves overall model performance, while selectively tuning the most critical layers significantly enhances fine-tuning efficiency with minimal performance loss.

Diversity-grounded Channel Prototypical Learning for Out-of-Distribution Intent Detection

In the realm of task-oriented dialogue systems, a robust intent detection mechanism must effectively handle malformed utterances encountered in real-world scenarios. This study presents a novel fine-tuning framework for large language models (LLMs) aimed at enhancing in-distribution (ID) intent classification and out-of-distribution (OOD) intent detection, which utilizes semantic matching with prototypes derived from ID class names. By harnessing the highly distinguishable representations of LLMs, we construct semantic prototypes for each ID class using a diversity-grounded prompt tuning approach. We rigorously test our framework in a challenging OOD context, where ID and OOD classes are semantically close yet distinct, referred to as \emph{near} OOD detection. For a thorough assessment, we benchmark our method against the prevalent fine-tuning approaches. The experimental findings reveal that our method demonstrates superior performance in both few-shot ID intent classification and near-OOD intent detection tasks.

MegaFake: A Theory-Driven Dataset of Fake News Generated by Large Language Models

The advent of large language models (LLMs) has revolutionized online content creation, making it much easier to generate high-quality fake news. This misuse threatens the integrity of our digital environment and ethical standards. Therefore, understanding the motivations and mechanisms behind LLM-generated fake news is crucial. In this study, we analyze the creation of fake news from a social psychology perspective and develop a comprehensive LLM-based theoretical framework, LLM-Fake Theory. We introduce a novel pipeline that automates the generation of fake news using LLMs, thereby eliminating the need for manual annotation. Utilizing this pipeline, we create a theoretically informed Machine-generated Fake news dataset, MegaFake, derived from the GossipCop dataset. We conduct comprehensive analyses to evaluate our MegaFake dataset. We believe that our dataset and insights will provide valuable contributions to future research focused on the detection and governance of fake news in the era of LLMs.

Continual Dialogue State Tracking via Reason-of-Select Distillation

An ideal dialogue system requires continuous skill acquisition and adaptation to new tasks while retaining prior knowledge. Dialogue State Tracking (DST), vital in these systems, often involves learning new services and confronting catastrophic forgetting, along with a critical capability loss termed the "Value Selection Quandary." To address these challenges, we introduce the Reason-of-Select (RoS) distillation method by enhancing smaller models with a novel 'meta-reasoning' capability. Meta-reasoning employs an enhanced multi-domain perspective, combining fragments of meta-knowledge from domain-specific dialogues during continual learning. This transcends traditional single-perspective reasoning. The domain bootstrapping process enhances the model's ability to dissect intricate dialogues from multiple possible values. Its domain-agnostic property aligns data distribution across different domains, effectively mitigating forgetting. Additionally, two novel improvements, "multi-value resolution" strategy and Semantic Contrastive Reasoning Selection method, significantly enhance RoS by generating DST-specific selection chains and mitigating hallucinations in teachers' reasoning, ensuring effective and reliable knowledge transfer. Extensive experiments validate the exceptional performance and robust generalization capabilities of our method. The source code is provided for reproducibility.

TaSL: Task Skill Localization and Consolidation for Language Model Continual Learning

Language model continual learning (CL) has recently garnered significant interest due to its potential to adapt large language models (LLMs) to dynamic real-world environments without re-training. A key challenge in this field is catastrophic forgetting, where models lose previously acquired knowledge when learning new tasks. Existing methods commonly employ multiple parameter-efficient fine-tuning (PEFT) blocks to acquire task-specific knowledge for each task, but these approaches lack efficiency and overlook the potential for knowledge transfer through task interaction. In this paper, we present a novel CL framework for language models called Task Skill Localization and Consolidation (TaSL), which enhances knowledge transfer without relying on memory replay. TaSL first divides the model into `skill units' based on parameter dependencies, enabling more granular control. It then employs a novel group-wise skill localization technique to identify the importance distribution of skill units for a new task. By comparing this importance distribution with those from previous tasks, we implement a fine-grained skill consolidation strategy that retains task-specific knowledge, thereby preventing forgetting, and updates task-shared knowledge, which facilitates bi-directional knowledge transfer. As a result, TaSL achieves a superior balance between retaining previous knowledge and excelling in new tasks. TaSL also shows strong generalizability, suitable for general models and customizable for PEFT methods like LoRA. Additionally, it demonstrates notable extensibility, allowing integration with memory replay to further enhance performance. Extensive experiments on two CL benchmarks, with varying model sizes (from 220M to 7B), demonstrate the effectiveness of TaSL and its variants across different settings.

Towards LLM-driven Dialogue State Tracking

Dialogue State Tracking (DST) is of paramount importance in ensuring accurate tracking of user goals and system actions within task-oriented dialogue systems. The emergence of large language models (LLMs) such as GPT3 and ChatGPT has sparked considerable interest in assessing their efficacy across diverse applications. In this study, we conduct an initial examination of ChatGPT's capabilities in DST. Our evaluation uncovers the exceptional performance of ChatGPT in this task, offering valuable insights to researchers regarding its capabilities and providing useful directions for designing and enhancing dialogue systems. Despite its impressive performance, ChatGPT has significant limitations including its closed-source nature, request restrictions, raising data privacy concerns, and lacking local deployment capabilities. To address these concerns, we present LDST, an LLM-driven DST framework based on smaller, open-source foundation models. By utilizing a novel domain-slot instruction tuning method, LDST achieves performance on par with ChatGPT. Comprehensive evaluations across three distinct experimental settings, we find that LDST exhibits remarkable performance improvements in both zero-shot and few-shot setting compared to previous SOTA methods. The source code is provided for reproducibility.

How Good Are Large Language Models at Out-of-Distribution Detection?

Out-of-distribution (OOD) detection plays a vital role in enhancing the reliability of machine learning (ML) models. The emergence of large language models (LLMs) has catalyzed a paradigm shift within the ML community, showcasing their exceptional capabilities across diverse natural language processing tasks. While existing research has probed OOD detection with relative small-scale Transformers like BERT, RoBERTa and GPT-2, the stark differences in scales, pre-training objectives, and inference paradigms call into question the applicability of these findings to LLMs. This paper embarks on a pioneering empirical investigation of OOD detection in the domain of LLMs, focusing on LLaMA series ranging from 7B to 65B in size. We thoroughly evaluate commonly-used OOD detectors, scrutinizing their performance in both zero-grad and fine-tuning scenarios. Notably, we alter previous discriminative in-distribution fine-tuning into generative fine-tuning, aligning the pre-training objective of LLMs with downstream tasks. Our findings unveil that a simple cosine distance OOD detector demonstrates superior efficacy, outperforming other OOD detectors. We provide an intriguing explanation for this phenomenon by highlighting the isotropic nature of the embedding spaces of LLMs, which distinctly contrasts with the anisotropic property observed in smaller BERT family models. The new insight enhances our understanding of how LLMs detect OOD data, thereby enhancing their adaptability and reliability in dynamic environments.

Hypernetwork-based Personalized Federated Learning for Multi-Institutional CT Imaging

Computed tomography (CT) is of great importance in clinical practice due to its powerful ability to provide patients' anatomical information without any invasive inspection, but its potential radiation risk is raising people's concerns. Deep learning-based methods are considered promising in CT reconstruction, but these network models are usually trained with the measured data obtained from specific scanning protocol and need to centralizedly collect large amounts of data, which will lead to serious data domain shift, and privacy concerns. To relieve these problems, in this paper, we propose a hypernetwork-based federated learning method for personalized CT imaging, dubbed as HyperFed. The basic assumption of HyperFed is that the optimization problem for each institution can be divided into two parts: the local data adaption problem and the global CT imaging problem, which are implemented by an institution-specific hypernetwork and a global-sharing imaging network, respectively. The purpose of global-sharing imaging network is to learn stable and effective common features from different institutions. The institution-specific hypernetwork is carefully designed to obtain hyperparameters to condition the global-sharing imaging network for personalized local CT reconstruction. Experiments show that HyperFed achieves competitive performance in CT reconstruction compared with several other state-of-the-art methods. It is believed as a promising direction to improve CT imaging quality and achieve personalized demands of different institutions or scanners without privacy data sharing. The codes will be released at https://github.com/Zi-YuanYang/HyperFed.

One Network to Solve Them All: A Sequential Multi-Task Joint Learning Network Framework for MR Imaging Pipeline

Magnetic resonance imaging (MRI) acquisition, reconstruction, and segmentation are usually processed independently in the conventional practice of MRI workflow. It is easy to notice that there are significant relevances among these tasks and this procedure artificially cuts off these potential connections, which may lead to losing clinically important information for the final diagnosis. To involve these potential relations for further performance improvement, a sequential multi-task joint learning network model is proposed to train a combined end-to-end pipeline in a differentiable way, aiming at exploring the mutual influence among those tasks simultaneously. Our design consists of three cascaded modules: 1) deep sampling pattern learning module optimizes the $k$-space sampling pattern with predetermined sampling rate; 2) deep reconstruction module is dedicated to reconstructing MR images from the undersampled data using the learned sampling pattern; 3) deep segmentation module encodes MR images reconstructed from the previous module to segment the interested tissues. The proposed model retrieves the latently interactive and cyclic relations among those tasks, from which each task will be mutually beneficial. The proposed framework is verified on MRB dataset, which achieves superior performance on other SOTA methods in terms of both reconstruction and segmentation.

IDOL-Net: An Interactive Dual-Domain Parallel Network for CT Metal Artifact Reduction

Due to the presence of metallic implants, the imaging quality of computed tomography (CT) would be heavily degraded. With the rapid development of deep learning, several network models have been proposed for metal artifact reduction (MAR). Since the dual-domain MAR methods can leverage the hybrid information from both sinogram and image domains, they have significantly improved the performance compared to single-domain methods. However,current dual-domain methods usually operate on both domains in a specific order, which implicitly imposes a certain priority prior into MAR and may ignore the latent information interaction between both domains. To address this problem, in this paper, we propose a novel interactive dualdomain parallel network for CT MAR, dubbed as IDOLNet. Different from existing dual-domain methods, the proposed IDOL-Net is composed of two modules. The disentanglement module is utilized to generate high-quality prior sinogram and image as the complementary inputs. The follow-up refinement module consists of two parallel and interactive branches that simultaneously operate on image and sinogram domain, fully exploiting the latent information interaction between both domains. The simulated and clinical results demonstrate that the proposed IDOL-Net outperforms several state-of-the-art models in both qualitative and quantitative aspects.