The rapid advancement of face forgery techniques has introduced a growing variety of forgeries. Incremental Face Forgery Detection (IFFD), involving gradually adding new forgery data to fine-tune the previously trained model, has been introduced as a promising strategy to deal with evolving forgery methods. However, a naively trained IFFD model is prone to catastrophic forgetting when new forgeries are integrated, as treating all forgeries as a single ''Fake" class in the Real/Fake classification can cause different forgery types overriding one another, thereby resulting in the forgetting of unique characteristics from earlier tasks and limiting the model's effectiveness in learning forgery specificity and generality. In this paper, we propose to stack the latent feature distributions of previous and new tasks brick by brick, $\textit{i.e.}$, achieving $\textbf{aligned feature isolation}$. In this manner, we aim to preserve learned forgery information and accumulate new knowledge by minimizing distribution overriding, thereby mitigating catastrophic forgetting. To achieve this, we first introduce Sparse Uniform Replay (SUR) to obtain the representative subsets that could be treated as the uniformly sparse versions of the previous global distributions. We then propose a Latent-space Incremental Detector (LID) that leverages SUR data to isolate and align distributions. For evaluation, we construct a more advanced and comprehensive benchmark tailored for IFFD. The leading experimental results validate the superiority of our method.
Large Language Models for Code (LLMs4Code) have been found to exhibit outstanding performance in the software engineering domain, especially the remarkable performance in coding tasks. However, even the most advanced LLMs4Code can inevitably contain incorrect or outdated code knowledge. Due to the high cost of training LLMs4Code, it is impractical to re-train the models for fixing these problematic code knowledge. Model editing is a new technical field for effectively and efficiently correcting erroneous knowledge in LLMs, where various model editing techniques and benchmarks have been proposed recently. Despite that, a comprehensive study that thoroughly compares and analyzes the performance of the state-of-the-art model editing techniques for adapting the knowledge within LLMs4Code across various code-related tasks is notably absent. To bridge this gap, we perform the first systematic study on applying state-of-the-art model editing approaches to repair the inaccuracy of LLMs4Code. To that end, we introduce a benchmark named CLMEEval, which consists of two datasets, i.e., CoNaLa-Edit (CNLE) with 21K+ code generation samples and CodeSearchNet-Edit (CSNE) with 16K+ code summarization samples. With the help of CLMEEval, we evaluate six advanced model editing techniques on three LLMs4Code: CodeLlama (7B), CodeQwen1.5 (7B), and Stable-Code (3B). Our findings include that the external memorization-based GRACE approach achieves the best knowledge editing effectiveness and specificity (the editing does not influence untargeted knowledge), while generalization (whether the editing can generalize to other semantically-identical inputs) is a universal challenge for existing techniques. Furthermore, building on in-depth case analysis, we introduce an enhanced version of GRACE called A-GRACE, which incorporates contrastive learning to better capture the semantics of the inputs.
Colorectal cancer (CRC) is one of the few cancers that have an established dysplasia-carcinoma sequence that benefits from screening. Everyone over 50 years of age in Canada is eligible for CRC screening. About 20\% of those people will undergo a biopsy for a pre-neoplastic polyp and, in many cases, multiple polyps. As such, these polyp biopsies make up the bulk of a pathologist's workload. Developing an efficient computational model to help screen these polyp biopsies can improve the pathologist's workflow and help guide their attention to critical areas on the slide. DL models face significant challenges in computational pathology (CPath) because of the gigapixel image size of whole-slide images and the scarcity of detailed annotated datasets. It is, therefore, crucial to leverage self-supervised learning (SSL) methods to alleviate the burden and cost of data annotation. However, current research lacks methods to apply SSL frameworks to analyze pathology data effectively. This paper aims to propose an optimized Barlow Twins framework for colorectal polyps screening. We adapt its hyperparameters, augmentation strategy and encoder to the specificity of the pathology data to enhance performance. Additionally, we investigate the best Field of View (FoV) for colorectal polyps screening and propose a new benchmark dataset for CRC screening, made of four types of colorectal polyps and normal tissue, by performing downstream tasking on MHIST and NCT-CRC-7K datasets. Furthermore, we show that the SSL representations are more meaningful and qualitative than the supervised ones and that Barlow Twins benefits from the Swin Transformer when applied to pathology data. Codes are avaialble from https://github.com/AtlasAnalyticsLab/PathBT.
In Human Activity Recognition (HAR), understanding the intricacy of body movements within high-risk applications is essential. This study uses SHapley Additive exPlanations (SHAP) to explain the decision-making process of Graph Convolution Networks (GCNs) when classifying activities with skeleton data. We employ SHAP to explain two real-world datasets: one for cerebral palsy (CP) classification and the widely used NTU RGB+D 60 action recognition dataset. To test the explanation, we introduce a novel perturbation approach that modifies the model's edge importance matrix, allowing us to evaluate the impact of specific body key points on prediction outcomes. To assess the fidelity of our explanations, we employ informed perturbation, targeting body key points identified as important by SHAP and comparing them against random perturbation as a control condition. This perturbation enables a judgment on whether the body key points are truly influential or non-influential based on the SHAP values. Results on both datasets show that body key points identified as important through SHAP have the largest influence on the accuracy, specificity, and sensitivity metrics. Our findings highlight that SHAP can provide granular insights into the input feature contribution to the prediction outcome of GCNs in HAR tasks. This demonstrates the potential for more interpretable and trustworthy models in high-stakes applications like healthcare or rehabilitation.
Existing multi-modal image fusion methods fail to address the compound degradations presented in source images, resulting in fusion images plagued by noise, color bias, improper exposure, \textit{etc}. Additionally, these methods often overlook the specificity of foreground objects, weakening the salience of the objects of interest within the fused images. To address these challenges, this study proposes a novel interactive multi-modal image fusion framework based on the text-modulated diffusion model, called Text-DiFuse. First, this framework integrates feature-level information integration into the diffusion process, allowing adaptive degradation removal and multi-modal information fusion. This is the first attempt to deeply and explicitly embed information fusion within the diffusion process, effectively addressing compound degradation in image fusion. Second, by embedding the combination of the text and zero-shot location model into the diffusion fusion process, a text-controlled fusion re-modulation strategy is developed. This enables user-customized text control to improve fusion performance and highlight foreground objects in the fused images. Extensive experiments on diverse public datasets show that our Text-DiFuse achieves state-of-the-art fusion performance across various scenarios with complex degradation. Moreover, the semantic segmentation experiment validates the significant enhancement in semantic performance achieved by our text-controlled fusion re-modulation strategy. The code is publicly available at https://github.com/Leiii-Cao/Text-DiFuse.
Consistency models are a new class of models that generate images by directly mapping noise to data, allowing for one-step generation and significantly accelerating the sampling process. However, their robustness against adversarial attacks has not yet been thoroughly investigated. In this work, we conduct the first study on the vulnerability of consistency models to backdoor attacks. While previous research has explored backdoor attacks on diffusion models, those studies have primarily focused on conventional diffusion models, employing a customized backdoor training process and objective, whereas consistency models have distinct training processes and objectives. Our proposed framework demonstrates the vulnerability of consistency models to backdoor attacks. During image generation, poisoned consistency models produce images with a Fr\'echet Inception Distance (FID) comparable to that of a clean model when sampling from Gaussian noise. However, once the trigger is activated, they generate backdoor target images. We explore various trigger and target configurations to evaluate the vulnerability of consistency models, including the use of random noise as a trigger. This novel trigger is visually inconspicuous, more challenging to detect, and aligns well with the sampling process of consistency models. Across all configurations, our framework successfully compromises the consistency models while maintaining high utility and specificity. We also examine the stealthiness of our proposed attack, which is attributed to the unique properties of consistency models and the elusive nature of the Gaussian noise trigger. Our code is available at \href{https://github.com/chengenw/backdoorCM}{https://github.com/chengenw/backdoorCM}.
Control system design is a crucial aspect of modern engineering with far-reaching applications across diverse sectors including aerospace, automotive systems, power grids, and robotics. Despite advances made by Large Language Models (LLMs) in various domains, their application in control system design remains limited due to the complexity and specificity of control theory. To bridge this gap, we introduce ControlAgent, a new paradigm that automates control system design via novel integration of LLM agents and control-oriented domain expertise. ControlAgent encodes expert control knowledge and emulates human iterative design processes by gradually tuning controller parameters to meet user-specified requirements for stability, performance, and robustness. ControlAgent integrates multiple collaborative LLM agents, including a central agent responsible for task distribution and task-specific agents dedicated to detailed controller design for various types of systems and requirements. ControlAgent also employs a Python computation agent that performs complex calculations and controller evaluations based on standard design information provided by task-specified LLM agents. Combined with a history and feedback module, the task-specific LLM agents iteratively refine controller parameters based on real-time feedback from prior designs. Overall, ControlAgent mimics the design processes used by (human) practicing engineers, but removes all the human efforts and can be run in a fully automated way to give end-to-end solutions for control system design with user-specified requirements. To validate ControlAgent's effectiveness, we develop ControlEval, an evaluation dataset that comprises 500 control tasks with various specific design goals. The effectiveness of ControlAgent is demonstrated via extensive comparative evaluations between LLM-based and traditional human-involved toolbox-based baselines.
Abnormalities in the gastrointestinal tract significantly influence the patient's health and require a timely diagnosis for effective treatment. With such consideration, an effective automatic classification of these abnormalities from a video capsule endoscopy (VCE) frame is crucial for improvement in diagnostic workflows. The work presents the process of developing and evaluating a novel model designed to classify gastrointestinal anomalies from a VCE video frame. Integration of Omni Dimensional Gated Attention (OGA) mechanism and Wavelet transformation techniques into the model's architecture allowed the model to focus on the most critical areas in the endoscopy images, reducing noise and irrelevant features. This is particularly advantageous in capsule endoscopy, where images often contain a high degree of variability in texture and color. Wavelet transformations contributed by efficiently capturing spatial and frequency-domain information, improving feature extraction, especially for detecting subtle features from the VCE frames. Furthermore, the features extracted from the Stationary Wavelet Transform and Discrete Wavelet Transform are concatenated channel-wise to capture multiscale features, which are essential for detecting polyps, ulcerations, and bleeding. This approach improves classification accuracy on imbalanced capsule endoscopy datasets. The proposed model achieved 92.76% and 91.19% as training and validation accuracies respectively. At the same time, Training and Validation losses are 0.2057 and 0.2700. The proposed model achieved a Balanced Accuracy of 94.81%, AUC of 87.49%, F1-score of 91.11%, precision of 91.17%, recall of 91.19% and specificity of 98.44%. Additionally, the model's performance is benchmarked against two base models, VGG16 and ResNet50, demonstrating its enhanced ability to identify and classify a range of gastrointestinal abnormalities accurately.
Pre-trained vision-language models like CLIP have remarkably adapted to various downstream tasks. Nonetheless, their performance heavily depends on the specificity of the input text prompts, which requires skillful prompt template engineering. Instead, current approaches to prompt optimization learn the prompts through gradient descent, where the prompts are treated as adjustable parameters. However, these methods tend to lead to overfitting of the base classes seen during training and produce prompts that are no longer understandable by humans. This paper introduces a simple but interpretable prompt optimizer (IPO), that utilizes large language models (LLMs) to generate textual prompts dynamically. We introduce a Prompt Optimization Prompt that not only guides LLMs in creating effective prompts but also stores past prompts with their performance metrics, providing rich in-context information. Additionally, we incorporate a large multimodal model (LMM) to condition on visual content by generating image descriptions, which enhance the interaction between textual and visual modalities. This allows for thae creation of dataset-specific prompts that improve generalization performance, while maintaining human comprehension. Extensive testing across 11 datasets reveals that IPO not only improves the accuracy of existing gradient-descent-based prompt learning methods but also considerably enhances the interpretability of the generated prompts. By leveraging the strengths of LLMs, our approach ensures that the prompts remain human-understandable, thereby facilitating better transparency and oversight for vision-language models.
Large language models (LLMs) have shown impressive potential in helping with numerous medical challenges. Deploying LLMs in high-stakes applications such as medicine, however, brings in many concerns. One major area of concern relates to biased behaviors of LLMs in medical applications, leading to unfair treatment of individuals. To pave the way for the responsible and impactful deployment of Med LLMs, rigorous evaluation is a key prerequisite. Due to the huge complexity and variability of different medical scenarios, existing work in this domain has primarily relied on using manually crafted datasets for bias evaluation. In this study, we present a new method to scale up such bias evaluations by automatically generating test cases based on rigorous medical evidence. We specifically target the challenges of a) domain-specificity of bias characterization, b) hallucinating while generating the test cases, and c) various dependencies between the health outcomes and sensitive attributes. To that end, we offer new methods to address these challenges integrated with our generative pipeline, using medical knowledge graphs, medical ontologies, and customized general LLM evaluation frameworks in our method. Through a series of extensive experiments, we show that the test cases generated by our proposed method can effectively reveal bias patterns in Med LLMs at larger and more flexible scales than human-crafted datasets. We publish a large bias evaluation dataset using our pipeline, which is dedicated to a few medical case studies. A live demo of our application for vignette generation is available at https://vignette.streamlit.app. Our code is also available at https://github.com/healthylaife/autofair.
Political debates are a peculiar type of political discourse, in which candidates directly confront one another, addressing not only the the moderator's questions, but also their opponent's statements, as well as the concerns of voters from both parties and undecided voters. Therefore, language is adjusted to meet specific expectations and achieve persuasion. We analyse how the language of Trump and Harris during the debate (September 10th 2024) differs in relation to the following semantic and pragmatic features, for which we formulated targeted hypotheses: framing values and ideology, appealing to emotion, using words with different degrees of concreteness and specificity, addressing others through singular or plural pronouns. Our findings include: differences in the use of figurative frames (Harris often framing issues around recovery and empowerment, Trump often focused on crisis and decline); similar use of emotional language, with Trump showing a slight higher tendency toward negativity and toward less subjective language compared to Harris; no significant difference in the specificity of candidates' responses; similar use of abstract language, with Trump showing more variability than Harris, depending on the subject discussed; differences in addressing the opponent, with Trump not mentioning Harris by name, while Harris referring to Trump frequently; different uses of pronouns, with Harris using both singular and plural pronouns equally, while Trump using more singular pronouns. The results are discussed in relation to previous literature on Red and Blue language, which refers to distinct linguistic patterns associated with conservative (Red) and liberal (Blue) political ideologies.
Counterspeech is a targeted response to counteract and challenge abusive or hateful content. It effectively curbs the spread of hatred and fosters constructive online communication. Previous studies have proposed different strategies for automatically generated counterspeech. Evaluations, however, focus on relevance, surface form, and other shallow linguistic characteristics. This paper investigates the human likeness of AI-generated counterspeech, a critical factor influencing effectiveness. We implement and evaluate several LLM-based generation strategies, and discover that AI-generated and human-written counterspeech can be easily distinguished by both simple classifiers and humans. Further, we reveal differences in linguistic characteristics, politeness, and specificity. The dataset used in this study is publicly available for further research.
Complex systems can undergo critical transitions, where slowly changing environmental conditions trigger a sudden shift to a new, potentially catastrophic state. Early warning signals for these events are crucial for decision-making in fields such as ecology, biology and climate science. Generic early warning signals motivated by dynamical systems theory have had mixed success on real noisy data. More recent studies found that deep learning classifiers trained on synthetic data could improve performance. However, neither of these methods take advantage of historical, system-specific data. Here, we introduce an approach that trains machine learning classifiers directly on surrogate data of past transitions, namely surrogate data-based machine learning (SDML). The approach provides early warning signals in empirical and experimental data from geology, climatology, sociology, and cardiology with higher sensitivity and specificity than two widely used generic early warning signals -- variance and lag-1 autocorrelation. Since the approach is trained directly on surrogates of historical data, it is not bound by the restricting assumption of a local bifurcation like previous methods. This system-specific approach can contribute to improved early warning signals to help humans better prepare for or avoid undesirable critical transitions.
In medical image synthesis, the precision of localized structural details is crucial, particularly when addressing specific clinical requirements such as the identification and measurement of fine structures. Traditional methods for image translation and synthesis are generally optimized for global image reconstruction but often fall short in providing the finesse required for detailed local analysis. This study represents a step toward addressing this challenge by introducing a novel anatomical feature-prioritized (AFP) loss function into the synthesis process. This method enhances reconstruction by focusing on clinically significant structures, utilizing features from a pre-trained model designed for a specific downstream task, such as the segmentation of particular anatomical regions. The AFP loss function can replace or complement global reconstruction methods, ensuring a balanced emphasis on both global image fidelity and local structural details. Various implementations of this loss function are explored, including its integration into different synthesis networks such as GAN-based and CNN-based models. Our approach is applied and evaluated in two contexts: lung MR to CT translation, focusing on high-quality reconstruction of bronchial structures, using a private dataset; and pelvis MR to CT synthesis, targeting the accurate representation of organs and muscles, utilizing a public dataset from the Synthrad2023 challenge. We leverage embeddings from pre-trained segmentation models specific to these anatomical regions to demonstrate the capability of the AFP loss to prioritize and accurately reconstruct essential features. This tailored approach shows promising potential for enhancing the specificity and practicality of medical image synthesis in clinical applications.
Epileptic seizures cause abnormal brain activity, and their unpredictability can lead to accidents, underscoring the need for long-term seizure prediction. Although seizures can be predicted by analyzing electroencephalogram (EEG) signals, existing methods often require too many electrode channels or larger models, limiting mobile usability. This paper introduces a SlimSeiz framework that utilizes adaptive channel selection with a lightweight neural network model. SlimSeiz operates in two states: the first stage selects the optimal channel set for seizure prediction using machine learning algorithms, and the second stage employs a lightweight neural network based on convolution and Mamba for prediction. On the Children's Hospital Boston-MIT (CHB-MIT) EEG dataset, SlimSeiz can reduce channels from 22 to 8 while achieving a satisfactory result of 94.8% accuracy, 95.5% sensitivity, and 94.0% specificity with only 21.2K model parameters, matching or outperforming larger models' performance. We also validate SlimSeiz on a new EEG dataset, SRH-LEI, collected from Shanghai Renji Hospital, demonstrating its effectiveness across different patients. The code and SRH-LEI dataset are available at https://github.com/guoruilu/SlimSeiz.
Evaluating the creativity of large language models (LLMs) in story writing is difficult because LLM-generated stories could seemingly look creative but be very similar to some existing stories in their huge and proprietary training corpus. To overcome this challenge, we introduce a novel benchmark dataset with varying levels of prompt specificity: CS4 ($\mathbf{C}$omparing the $\mathbf{S}$kill of $\mathbf{C}$reating $\mathbf{S}$tories by $\mathbf{C}$ontrolling the $\mathbf{S}$ynthesized $\mathbf{C}$onstraint $\mathbf{S}$pecificity). By increasing the number of requirements/constraints in the prompt, we can increase the prompt specificity and hinder LLMs from retelling high-quality narratives in their training data. Consequently, CS4 empowers us to indirectly measure the LLMs' creativity without human annotations. Our experiments on LLaMA, Gemma, and Mistral not only highlight the creativity challenges LLMs face when dealing with highly specific prompts but also reveal that different LLMs perform very differently under different numbers of constraints and achieve different balances between the model's instruction-following ability and narrative coherence. Additionally, our experiments on OLMo suggest that Learning from Human Feedback (LHF) can help LLMs select better stories from their training data but has limited influence in boosting LLMs' ability to produce creative stories that are unseen in the training corpora. The benchmark is released at https://github.com/anirudhlakkaraju/cs4_benchmark.
With the rapid advancement of Natural Language Processing in recent years, numerous studies have shown that generic summaries generated by Large Language Models (LLMs) can sometimes surpass those annotated by experts, such as journalists, according to human evaluations. However, there is limited research on whether these generic summaries meet the individual needs of ordinary people. The biggest obstacle is the lack of human-annotated datasets from the general public. Existing work on personalized summarization often relies on pseudo datasets created from generic summarization datasets or controllable tasks that focus on specific named entities or other aspects, such as the length and specificity of generated summaries, collected from hypothetical tasks without the annotators' initiative. To bridge this gap, we propose a high-quality, personalized, manually annotated abstractive summarization dataset called PersonalSum. This dataset is the first to investigate whether the focus of public readers differs from the generic summaries generated by LLMs. It includes user profiles, personalized summaries accompanied by source sentences from given articles, and machine-generated generic summaries along with their sources. We investigate several personal signals - entities/topics, plot, and structure of articles - that may affect the generation of personalized summaries using LLMs in a few-shot in-context learning scenario. Our preliminary results and analysis indicate that entities/topics are merely one of the key factors that impact the diverse preferences of users, and personalized summarization remains a significant challenge for existing LLMs.
Large Language Models trained on web-scale text acquire language generation abilities that can solve a wide range of tasks, particularly when task knowledge is refined into the generative prior using in-context examples. However, spurious features learned from noisy data hinder their generalizability. Supervised finetuning can introduce task specificity, but introduce data inefficiency. Prior studies indicate that (i) noisy neural circuitries coexist with generalizable ones within LLMs, and (ii) finetuning typically enhances (or suppresses) existing abilities without introducing newer ones. Building upon these, we propose TaRot, a novel method for task adaptation. TaRot intervenes in the neural circuitries using learnable rotation matrices that are optimized using Bayesian Optimization, on labelled samples in the order of standard few-shot prompting examples. Experiments on multiple classification and generation tasks using LLMs of varying sizes reveal the efficacy of TaRot, improving upon both zero- as well as few-shot performance, with average improvements (across models and tasks) of 23.81% and 11.15%, respectively. The source code is available at https://github.com/joykirat18/TaRot
In this work, we propose Erasure of Language Memory (ELM), an approach for concept-level unlearning built on the principle of matching the distribution defined by an introspective classifier. Our key insight is that effective unlearning should leverage the model's ability to evaluate its own knowledge, using the model itself as a classifier to identify and reduce the likelihood of generating content related to undesired concepts. ELM applies this framework to create targeted low-rank updates that reduce generation probabilities for concept-specific content while preserving the model's broader capabilities. We demonstrate ELM's efficacy on biosecurity, cybersecurity, and literary domain erasure tasks. Comparative analysis shows that ELM achieves superior performance across key metrics, including near-random scores on erased topic assessments, maintained coherence in text generation, preserved accuracy on unrelated benchmarks, and robustness under adversarial attacks. Our code, data, and trained models are available at https://elm.baulab.info
As Large Language Models (LLMs) grow increasingly adept at managing complex tasks, the evaluation set must keep pace with these advancements to ensure it remains sufficiently discriminative. Item Discrimination (ID) theory, which is widely used in educational assessment, measures the ability of individual test items to differentiate between high and low performers. Inspired by this theory, we propose an ID-induced prompt synthesis framework for evaluating LLMs to ensure the evaluation set can continually update and refine according to model abilities. Our data synthesis framework prioritizes both breadth and specificity. It can generate prompts that comprehensively evaluate the capabilities of LLMs while revealing meaningful performance differences between models, allowing for effective discrimination of their relative strengths and weaknesses across various tasks and domains. To produce high-quality data, we incorporate a self-correct mechanism into our generalization framework, and develop two models to predict prompt discrimination and difficulty score to facilitate our data synthesis framework, contributing valuable tools to evaluation data synthesis research. We apply our generated data to evaluate five SOTA models. Our data achieves an average score of 51.92, accompanied by a variance of 10.06. By contrast, previous works (i.e., SELF-INSTRUCT and WizardLM) obtain an average score exceeding 67, with a variance below 3.2. The results demonstrate that the data generated by our framework is more challenging and discriminative compared to previous works. We will release a dataset of over 3,000 carefully crafted prompts to facilitate evaluation research of LLMs.