Despite vision-language models' (VLMs) remarkable capabilities as versatile visual assistants, two substantial challenges persist within the existing VLM frameworks: (1) lacking task diversity in pretraining and visual instruction tuning, and (2) annotation error and bias in GPT-4 synthesized instruction tuning data. Both challenges lead to issues such as poor generalizability, hallucination, and catastrophic forgetting. To address these challenges, we construct Vision-Flan, the most diverse publicly available visual instruction tuning dataset to date, comprising 187 diverse tasks and 1,664,261 instances sourced from academic datasets, and each task is accompanied by an expert-written instruction. In addition, we propose a two-stage instruction tuning framework, in which VLMs are firstly finetuned on Vision-Flan and further tuned on GPT-4 synthesized data. We find this two-stage tuning framework significantly outperforms the traditional single-stage visual instruction tuning framework and achieves the state-of-the-art performance across a wide range of multi-modal evaluation benchmarks. Finally, we conduct in-depth analyses to understand visual instruction tuning and our findings reveal that: (1) GPT-4 synthesized data does not substantially enhance VLMs' capabilities but rather modulates the model's responses to human-preferred formats; (2) A minimal quantity (e.g., 1,000) of GPT-4 synthesized data can effectively align VLM responses with human-preference; (3) Visual instruction tuning mainly helps large-language models (LLMs) to understand visual features.
The ability to associate touch with other modalities has huge implications for humans and computational systems. However, multimodal learning with touch remains challenging due to the expensive data collection process and non-standardized sensor outputs. We introduce UniTouch, a unified tactile model for vision-based touch sensors connected to multiple modalities, including vision, language, and sound. We achieve this by aligning our UniTouch embeddings to pretrained image embeddings already associated with a variety of other modalities. We further propose learnable sensor-specific tokens, allowing the model to learn from a set of heterogeneous tactile sensors, all at the same time. UniTouch is capable of conducting various touch sensing tasks in the zero-shot setting, from robot grasping prediction to touch image question answering. To the best of our knowledge, UniTouch is the first to demonstrate such capabilities. Project page: https://cfeng16.github.io/UniTouch/
The rapid integration of Federated Learning (FL) into networking encompasses various aspects such as network management, quality of service, and cybersecurity while preserving data privacy. In this context, Decentralized Federated Learning (DFL) emerges as an innovative paradigm to train collaborative models, addressing the single point of failure limitation. However, the security and trustworthiness of FL and DFL are compromised by poisoning attacks, negatively impacting its performance. Existing defense mechanisms have been designed for centralized FL and they do not adequately exploit the particularities of DFL. Thus, this work introduces Sentinel, a defense strategy to counteract poisoning attacks in DFL. Sentinel leverages the accessibility of local data and defines a three-step aggregation protocol consisting of similarity filtering, bootstrap validation, and normalization to safeguard against malicious model updates. Sentinel has been evaluated with diverse datasets and various poisoning attack types and threat levels, improving the state-of-the-art performance against both untargeted and targeted poisoning attacks.
Large language models (LLMs), such as ChatGPT and GPT-4, are versatile and can solve different tasks due to their emergent ability and generalizability. However, LLMs sometimes lack domain-specific knowledge to perform tasks, which would also cause hallucination during inference. In some previous works, additional modules like graph neural networks (GNNs) are trained on retrieved knowledge from external knowledge bases, aiming to mitigate the problem of lacking domain-specific knowledge. However, incorporating additional modules: 1) would need retraining additional modules when encountering novel domains; 2) would become a bottleneck since LLMs' strong abilities are not fully utilized for retrieval. In this paper, we propose a paradigm, termed Knowledge Solver (KSL), to teach LLMs to search for essential knowledge from external knowledge bases by harnessing their own strong generalizability. Specifically, we design a simple yet effective prompt to transform retrieval into a multi-hop decision sequence, which empowers LLMs with searching knowledge ability in zero-shot manner. Additionally, KSL is able to provide complete retrieval paths and therefore increase explainability of LLMs' reasoning processes. We conduct experiments on three datasets: CommonsenseQA, OpenbookQA, and MedQA-USMLE, and found that our approach improves LLM baseline performance by a relatively large margin.
The expansion of the Internet-of-Things (IoT) paradigm is inevitable, but vulnerabilities of IoT devices to malware incidents have become an increasing concern. Recent research has shown that the integration of Reinforcement Learning with Moving Target Defense (MTD) mechanisms can enhance cybersecurity in IoT devices. Nevertheless, the numerous new malware attacks and the time that agents take to learn and select effective MTD techniques make this approach impractical for real-world IoT scenarios. To tackle this issue, this work presents CyberForce, a framework that employs Federated Reinforcement Learning (FRL) to collectively and privately determine suitable MTD techniques for mitigating diverse zero-day attacks. CyberForce integrates device fingerprinting and anomaly detection to reward or penalize MTD mechanisms chosen by an FRL-based agent. The framework has been evaluated in a federation consisting of ten devices of a real IoT platform. A pool of experiments with six malware samples affecting the devices has demonstrated that CyberForce can precisely learn optimum MTD mitigation strategies. When all clients are affected by all attacks, the FRL agent exhibits high accuracy and reduced training time when compared to a centralized RL agent. In cases where different clients experience distinct attacks, the CyberForce clients gain benefits through the transfer of knowledge from other clients and similar attack behavior. Additionally, CyberForce showcases notable robustness against data poisoning attacks.
Digitization increases business opportunities and the risk of companies being victims of devastating cyberattacks. Therefore, managing risk exposure and cybersecurity strategies is essential for digitized companies that want to survive in competitive markets. However, understanding company-specific risks and quantifying their associated costs is not trivial. Current approaches fail to provide individualized and quantitative monetary estimations of cybersecurity impacts. Due to limited resources and technical expertise, SMEs and even large companies are affected and struggle to quantify their cyberattack exposure. Therefore, novel approaches must be placed to support the understanding of the financial loss due to cyberattacks. This article introduces the Real Cyber Value at Risk (RCVaR), an economical approach for estimating cybersecurity costs using real-world information from public cybersecurity reports. RCVaR identifies the most significant cyber risk factors from various sources and combines their quantitative results to estimate specific cyberattacks costs for companies. Furthermore, RCVaR extends current methods to achieve cost and risk estimations based on historical real-world data instead of only probability-based simulations. The evaluation of the approach on unseen data shows the accuracy and efficiency of the RCVaR in predicting and managing cyber risks. Thus, it shows that the RCVaR is a valuable addition to cybersecurity planning and risk management processes.
In 2016, Google proposed Federated Learning (FL) as a novel paradigm to train Machine Learning (ML) models across the participants of a federation while preserving data privacy. Since its birth, Centralized FL (CFL) has been the most used approach, where a central entity aggregates participants' models to create a global one. However, CFL presents limitations such as communication bottlenecks, single point of failure, and reliance on a central server. Decentralized Federated Learning (DFL) addresses these issues by enabling decentralized model aggregation and minimizing dependency on a central entity. Despite these advances, current platforms training DFL models struggle with key issues such as managing heterogeneous federation network topologies. To overcome these challenges, this paper presents Fedstellar, a novel platform designed to train FL models in a decentralized, semi-decentralized, and centralized fashion across diverse federations of physical or virtualized devices. The Fedstellar implementation encompasses a web application with an interactive graphical interface, a controller for deploying federations of nodes using physical or virtual devices, and a core deployed on each device which provides the logic needed to train, aggregate, and communicate in the network. The effectiveness of the platform has been demonstrated in two scenarios: a physical deployment involving single-board devices such as Raspberry Pis for detecting cyberattacks, and a virtualized deployment comparing various FL approaches in a controlled environment using MNIST and CIFAR-10 datasets. In both scenarios, Fedstellar demonstrated consistent performance and adaptability, achieving F1 scores of 91%, 98%, and 91.2% using DFL for detecting cyberattacks and classifying MNIST and CIFAR-10, respectively, reducing training time by 32% compared to centralized approaches.
Manipulated videos often contain subtle inconsistencies between their visual and audio signals. We propose a video forensics method, based on anomaly detection, that can identify these inconsistencies, and that can be trained solely using real, unlabeled data. We train an autoregressive model to generate sequences of audio-visual features, using feature sets that capture the temporal synchronization between video frames and sound. At test time, we then flag videos that the model assigns low probability. Despite being trained entirely on real videos, our model obtains strong performance on the task of detecting manipulated speech videos. Project site: https://cfeng16.github.io/audio-visual-forensics
Wireless communication using fully passive metal reflectors is a promising technique for coverage expansion, signal enhancement, rank improvement and blind-zone compensation, thanks to its appealing features including zero energy consumption, ultra low cost, signaling- and maintenance-free, easy deployment and full compatibility with existing and future wireless systems. However, a prevalent understanding for reflection by metal plates is based on Snell's Law, i.e., signal can only be received when the observation angle equals to the incident angle, which is valid only when the electrical dimension of the metal plate is extremely large. In this paper, we rigorously derive a general reflection model that is applicable to metal reflectors of any size, any orientation, and any linear polarization. The derived model is given compactly in terms of the radar cross section (RCS) of the metal plate, as a function of its physical dimensions and orientation vectors, as well as the wave polarization and the wave deflection vector, i.e., the change of direction from the incident wave direction to the observation direction. Furthermore, experimental results based on actual field measurements are provided to validate the accuracy of our developed model and demonstrate the great potential of communications using metal reflectors.
Diabetic Retinopathy (DR) has become one of the leading causes of vision impairment in working-aged people and is a severe problem worldwide. However, most of the works ignored the ordinal information of labels. In this project, we propose a novel design MTCSNN, a Multi-task Clinical Siamese Neural Network for Diabetic Retinopathy severity prediction task. The novelty of this project is to utilize the ordinal information among labels and add a new regression task, which can help the model learn more discriminative feature embedding for fine-grained classification tasks. We perform comprehensive experiments over the RetinaMNIST, comparing MTCSNN with other models like ResNet-18, 34, 50. Our results indicate that MTCSNN outperforms the benchmark models in terms of AUC and accuracy on the test dataset.