Improving Medical Diagnostics with Vision-Language Models: Convex Hull-Based Uncertainty Analysis

In recent years, vision-language models (VLMs) have been applied to various fields, including healthcare, education, finance, and manufacturing, with remarkable performance. However, concerns remain regarding VLMs' consistency and uncertainty, particularly in critical applications such as healthcare, which demand a high level of trust and reliability. This paper proposes a novel approach to evaluate uncertainty in VLMs' responses using a convex hull approach on a healthcare application for Visual Question Answering (VQA). LLM-CXR model is selected as the medical VLM utilized to generate responses for a given prompt at different temperature settings, i.e., 0.001, 0.25, 0.50, 0.75, and 1.00. According to the results, the LLM-CXR VLM shows a high uncertainty at higher temperature settings. Experimental outcomes emphasize the importance of uncertainty in VLMs' responses, especially in healthcare applications.

Neural Networks Meet Elliptic Curve Cryptography: A Novel Approach to Secure Communication

In recent years, neural networks have been used to implement symmetric cryptographic functions for secure communications. Extending this domain, the proposed approach explores the application of asymmetric cryptography within a neural network framework to safeguard the exchange between two communicating entities, i.e., Alice and Bob, from an adversarial eavesdropper, i.e., Eve. It employs a set of five distinct cryptographic keys to examine the efficacy and robustness of communication security against eavesdropping attempts using the principles of elliptic curve cryptography. The experimental setup reveals that Alice and Bob achieve secure communication with negligible variation in security effectiveness across different curves. It is also designed to evaluate cryptographic resilience. Specifically, the loss metrics for Bob oscillate between 0 and 1 during encryption-decryption processes, indicating successful message comprehension post-encryption by Alice. The potential vulnerability with a decryption accuracy exceeds 60\%, where Eve experiences enhanced adversarial training, receiving twice the training iterations per batch compared to Alice and Bob.

Uncertainty Quantification in Large Language Models Through Convex Hull Analysis

Uncertainty quantification approaches have been more critical in large language models (LLMs), particularly high-risk applications requiring reliable outputs. However, traditional methods for uncertainty quantification, such as probabilistic models and ensemble techniques, face challenges when applied to the complex and high-dimensional nature of LLM-generated outputs. This study proposes a novel geometric approach to uncertainty quantification using convex hull analysis. The proposed method leverages the spatial properties of response embeddings to measure the dispersion and variability of model outputs. The prompts are categorized into three types, i.e., `easy', `moderate', and `confusing', to generate multiple responses using different LLMs at varying temperature settings. The responses are transformed into high-dimensional embeddings via a BERT model and subsequently projected into a two-dimensional space using Principal Component Analysis (PCA). The Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is utilized to cluster the embeddings and compute the convex hull for each selected cluster. The experimental results indicate that the uncertainty of the model for LLMs depends on the prompt complexity, the model, and the temperature setting.

Anomaly Detection in Power Markets and Systems

The widespread use of information and communication technology (ICT) over the course of the last decades has been a primary catalyst behind the digitalization of power systems. Meanwhile, as the utilization rate of the Internet of Things (IoT) continues to rise along with recent advancements in ICT, the need for secure and computationally efficient monitoring of critical infrastructures like the electrical grid and the agents that participate in it is growing. A cyber-physical system, such as the electrical grid, may experience anomalies for a number of different reasons. These may include physical defects, mistakes in measurement and communication, cyberattacks, and other similar occurrences. The goal of this study is to emphasize what the most common incidents are with power systems and to give an overview and classification of the most common ways to find problems, starting with the consumer/prosumer end working up to the primary power producers. In addition, this article aimed to discuss the methods and techniques, such as artificial intelligence (AI) that are used to identify anomalies in the power systems and markets.

Mitigating Attacks on Artificial Intelligence-based Spectrum Sensing for Cellular Network Signals

Cellular networks (LTE, 5G, and beyond) are dramatically growing with high demand from consumers and more promising than the other wireless networks with advanced telecommunication technologies. The main goal of these networks is to connect billions of devices, systems, and users with high-speed data transmission, high cell capacity, and low latency, as well as to support a wide range of new applications, such as virtual reality, metaverse, telehealth, online education, autonomous and flying vehicles, advanced manufacturing, and many more. To achieve these goals, spectrum sensing has been paid more attention, along with new approaches using artificial intelligence (AI) methods for spectrum management in cellular networks. This paper provides a vulnerability analysis of spectrum sensing approaches using AI-based semantic segmentation models for identifying cellular network signals under adversarial attacks with and without defensive distillation methods. The results showed that mitigation methods can significantly reduce the vulnerabilities of AI-based spectrum sensing models against adversarial attacks.

Hybrid AI-based Anomaly Detection Model using Phasor Measurement Unit Data

Over the last few decades, extensive use of information and communication technologies has been the main driver of the digitalization of power systems. Proper and secure monitoring of the critical grid infrastructure became an integral part of the modern power system. Using phasor measurement units (PMUs) to surveil the power system is one of the technologies that have a promising future. Increased frequency of measurements and smarter methods for data handling can improve the ability to reliably operate power grids. The increased cyber-physical interaction offers both benefits and drawbacks, where one of the drawbacks comes in the form of anomalies in the measurement data. The anomalies can be caused by both physical faults on the power grid, as well as disturbances, errors, and cyber attacks in the cyber layer. This paper aims to develop a hybrid AI-based model that is based on various methods such as Long Short Term Memory (LSTM), Convolutional Neural Network (CNN) and other relevant hybrid algorithms for anomaly detection in phasor measurement unit data. The dataset used within this research was acquired by the University of Texas, which consists of real data from grid measurements. In addition to the real data, false data that has been injected to produce anomalies has been analyzed. The impacts and mitigating methods to prevent such kind of anomalies are discussed.

Defensive Distillation based Adversarial Attacks Mitigation Method for Channel Estimation using Deep Learning Models in Next-Generation Wireless Networks

Future wireless networks (5G and beyond) are the vision of forthcoming cellular systems, connecting billions of devices and people together. In the last decades, cellular networks have been dramatically growth with advanced telecommunication technologies for high-speed data transmission, high cell capacity, and low latency. The main goal of those technologies is to support a wide range of new applications, such as virtual reality, metaverse, telehealth, online education, autonomous and flying vehicles, smart cities, smart grids, advanced manufacturing, and many more. The key motivation of NextG networks is to meet the high demand for those applications by improving and optimizing network functions. Artificial Intelligence (AI) has a high potential to achieve these requirements by being integrated in applications throughout all layers of the network. However, the security concerns on network functions of NextG using AI-based models, i.e., model poising, have not been investigated deeply. Therefore, it needs to design efficient mitigation techniques and secure solutions for NextG networks using AI-based methods. This paper proposes a comprehensive vulnerability analysis of deep learning (DL)-based channel estimation models trained with the dataset obtained from MATLAB's 5G toolbox for adversarial attacks and defensive distillation-based mitigation methods. The adversarial attacks produce faulty results by manipulating trained DL-based models for channel estimation in NextG networks, while making models more robust against any attacks through mitigation methods. This paper also presents the performance of the proposed defensive distillation mitigation method for each adversarial attack against the channel estimation model. The results indicated that the proposed mitigation method can defend the DL-based channel estimation models against adversarial attacks in NextG networks.

Homomorphic Encryption and Federated Learning based Privacy-Preserving CNN Training: COVID-19 Detection Use-Case

Medical data is often highly sensitive in terms of data privacy and security concerns. Federated learning, one type of machine learning techniques, has been started to use for the improvement of the privacy and security of medical data. In the federated learning, the training data is distributed across multiple machines, and the learning process is performed in a collaborative manner. There are several privacy attacks on deep learning (DL) models to get the sensitive information by attackers. Therefore, the DL model itself should be protected from the adversarial attack, especially for applications using medical data. One of the solutions for this problem is homomorphic encryption-based model protection from the adversary collaborator. This paper proposes a privacy-preserving federated learning algorithm for medical data using homomorphic encryption. The proposed algorithm uses a secure multi-party computation protocol to protect the deep learning model from the adversaries. In this study, the proposed algorithm using a real-world medical dataset is evaluated in terms of the model performance.

The Adversarial Security Mitigations of mmWave Beamforming Prediction Models using Defensive Distillation and Adversarial Retraining

The design of a security scheme for beamforming prediction is critical for next-generation wireless networks (5G, 6G, and beyond). However, there is no consensus about protecting the beamforming prediction using deep learning algorithms in these networks. This paper presents the security vulnerabilities in deep learning for beamforming prediction using deep neural networks (DNNs) in 6G wireless networks, which treats the beamforming prediction as a multi-output regression problem. It is indicated that the initial DNN model is vulnerable against adversarial attacks, such as Fast Gradient Sign Method (FGSM), Basic Iterative Method (BIM), Projected Gradient Descent (PGD), and Momentum Iterative Method (MIM), because the initial DNN model is sensitive to the perturbations of the adversarial samples of the training data. This study also offers two mitigation methods, such as adversarial training and defensive distillation, for adversarial attacks against artificial intelligence (AI)-based models used in the millimeter-wave (mmWave) beamforming prediction. Furthermore, the proposed scheme can be used in situations where the data are corrupted due to the adversarial examples in the training data. Experimental results show that the proposed methods effectively defend the DNN models against adversarial attacks in next-generation wireless networks.

Unreasonable Effectiveness of Last Hidden Layer Activations

In standard Deep Neural Network (DNN) based classifiers, the general convention is to omit the activation function in the last (output) layer and directly apply the softmax function on the logits to get the probability scores of each class. In this type of architectures, the loss value of the classifier against any output class is directly proportional to the difference between the final probability score and the label value of the associated class. Standard White-box adversarial evasion attacks, whether targeted or untargeted, mainly try to exploit the gradient of the model loss function to craft adversarial samples and fool the model. In this study, we show both mathematically and experimentally that using some widely known activation functions in the output layer of the model with high temperature values has the effect of zeroing out the gradients for both targeted and untargeted attack cases, preventing attackers from exploiting the model's loss function to craft adversarial samples. We've experimentally verified the efficacy of our approach on MNIST (Digit), CIFAR10 datasets. Detailed experiments confirmed that our approach substantially improves robustness against gradient-based targeted and untargeted attack threats. And, we showed that the increased non-linearity at the output layer has some additional benefits against some other attack methods like Deepfool attack.