Machine learning models are prone to memorizing sensitive data, making them vulnerable to membership inference attacks in which an adversary aims to guess if an input sample was used to train the model. In this paper, we show that prior work on membership inference attacks may severely underestimate the privacy risks by relying solely on training custom neural network classifiers to perform attacks and focusing only on the aggregate results over data samples, such as the attack accuracy. To overcome these limitations, we first propose to benchmark membership inference privacy risks by improving existing non-neural network based inference attacks and proposing a new inference attack method based on a modification of prediction entropy. We also propose benchmarks for defense mechanisms by accounting for adaptive adversaries with knowledge of the defense and also accounting for the trade-off between model accuracy and privacy risks. Using our benchmark attacks, we demonstrate that existing defense approaches are not as effective as previously reported. Next, we introduce a new approach for fine-grained privacy analysis by formulating and deriving a new metric called the privacy risk score. Our privacy risk score metric measures an individual sample's likelihood of being a training member, which allows an adversary to perform membership inference attacks with high confidence. We experimentally validate the effectiveness of the privacy risk score metric and demonstrate that the distribution of the privacy risk score across individual samples is heterogeneous. Finally, we perform an in-depth investigation for understanding why certain samples have high privacy risk scores, including correlations with model sensitivity, generalization error, and feature embeddings. Our work emphasizes the importance of a systematic and rigorous evaluation of privacy risks of machine learning models.

In order to increase productivity, capability, and data exploitation, numerous defense applications are experiencing an integration of state-of-the-art machine learning and AI into their architectures. Especially for defense applications, having a human analyst in the loop is of high interest due to quality control, accountability, and complex subject matter expertise not readily automated or replicated by AI. However, many applications are suffering from a very slow transition. This may be in large part due to lack of trust, usability, and productivity, especially when adapting to unforeseen classes and changes in mission context. Interactive machine learning is a newly emerging field in which machine learning implementations are trained, optimized, evaluated, and exploited through an intuitive human-computer interface. In this paper, we introduce interactive machine learning and explain its advantages and limitations within the context of defense applications. Furthermore, we address several of the shortcomings of interactive machine learning by discussing how cognitive feedback may inform features, data, and results in the state of the art. We define the three techniques by which cognitive feedback may be employed: self reporting, implicit cognitive feedback, and modeled cognitive feedback. The advantages and disadvantages of each technique are discussed.

Counterfactual prediction is a fundamental task in decision-making. G-computation is a method for estimating expected counterfactual outcomes under dynamic time-varying treatment strategies. Existing G-computation implementations have mostly employed classical regression models with limited capacity to capture complex temporal and nonlinear dependence structures. This paper introduces G-Net, a novel sequential deep learning framework for G-computation that can handle complex time series data while imposing minimal modeling assumptions and provide estimates of individual or population-level time varying treatment effects. We evaluate alternative G-Net implementations using realistically complex temporal simulated data obtained from CVSim, a mechanistic model of the cardiovascular system.

Artificial Intelligence (AI) in healthcare holds great potential to expand access to high-quality medical care, whilst reducing overall systemic costs. Despite hitting the headlines regularly and many publications of proofs-of-concept, certified products are failing to breakthrough to the clinic. AI in healthcare is a multi-party process with deep knowledge required in multiple individual domains. The lack of understanding of the specific challenges in the domain is, therefore, the major contributor to the failure to deliver on the big promises. Thus, we present a decision perspective framework, for the development of AI-driven biomedical products, from conception to market launch. Our framework highlights the risks, objectives and key results which are typically required to proceed through a three-phase process to the market launch of a validated medical AI product. We focus on issues related to Clinical validation, Regulatory affairs, Data strategy and Algorithmic development. The development process we propose for AI in healthcare software strongly diverges from modern consumer software development processes. We highlight the key time points to guide founders, investors and key stakeholders throughout their relevant part of the process. Our framework should be seen as a template for innovation frameworks, which can be used to coordinate team communications and responsibilities towards a reasonable product development roadmap, thus unlocking the potential of AI in medicine.

The ever evolving informatics technology has gradually bounded human and computer in a compact way. Understanding user behavior becomes a key enabler in many fields such as sedentary-related healthcare, human-computer interaction (HCI) and affective computing. Traditional sensor-based and vision-based user behavior analysis approaches are obtrusive in general, hindering their usage in realworld. Therefore, in this article, we first introduce WiFi signal as a new source instead of sensor and vision for unobtrusive user behaviors analysis. Then we design BeSense, a contactless behavior analysis system leveraging signal processing and computational intelligence over WiFi channel state information (CSI). We prototype BeSense on commodity low-cost WiFi devices and evaluate its performance in realworld environments. Experimental results have verified its effectiveness in recognizing user behaviors.

Despite the fact that the loss functions of deep neural networks are highly non-convex, gradient-based optimization algorithms converge to approximately the same performance from many random initial points. One thread of work has focused on explaining this phenomenon by characterizing the local curvature near critical points of the loss function, where the gradients are near zero, and demonstrating that neural network losses enjoy a no-bad-local-minima property and an abundance of saddle points. We report here that the methods used to find these putative critical points suffer from a bad local minima problem of their own: they often converge to or pass through regions where the gradient norm has a stationary point. We call these gradient-flat regions, since they arise when the gradient is approximately in the kernel of the Hessian, such that the loss is locally approximately linear, or flat, in the direction of the gradient. We describe how the presence of these regions necessitates care in both interpreting past results that claimed to find critical points of neural network losses and in designing second-order methods for optimizing neural networks.

For many applications, such as computing the expected value of different magnitudes, sampling from a known probability density function, the target density, is crucial but challenging through the inverse transform. In these cases, rejection and importance sampling require suitable proposal densities, which can be evaluated and sampled from efficiently. We will present a method based on normalizing flows, proposing a solution for the common problem of exploding reverse Kullback-Leibler divergence due to the target density having values of 0 in regions of the flow transformation. The performance of the method will be demonstrated using a multi-mode complex density function.

Recent neural network models for algorithmic tasks have led to significant improvements in extrapolation to sequences much longer than training, but it remains an outstanding problem that the performance still degrades for very long or adversarial sequences. We present alternative architectures and loss-terms to address these issues, and our testing of these approaches has not detected any remaining extrapolation errors within memory constraints. We focus on linear time algorithmic tasks including copy, parentheses parsing, and binary addition. First, activation binning was used to discretize the trained network in order to avoid computational drift from continuous operations, and a binning-based digital loss term was added to encourage discretizable representations. In addition, a localized differentiable memory (LDM) architecture, in contrast to distributed memory access, addressed remaining extrapolation errors and avoided unbounded growth of internal computational states. Previous work has found that algorithmic extrapolation issues can also be alleviated with approaches relying on program traces, but the current effort does not rely on such traces.

In this paper we present a novel approach to optimise tactical and strategic decision making in football (soccer). We model the game of football as a multi-stage game which is made up from a Bayesian game to model the pre-match decisions and a stochastic game to model the in-match state transitions and decisions. Using this formulation, we propose a method to predict the probability of game outcomes and the payoffs of team actions. Building upon this, we develop algorithms to optimise team formation and in-game tactics with different objectives. Empirical evaluation of our approach on real-world datasets from 760 matches shows that by using optimised tactics from our Bayesian and stochastic games, we can increase a team chances of winning by up to 16.1\% and 3.4\% respectively.

Deep neural networks have been widely adopted in modern reinforcement learning (RL) algorithms with great empirical successes in various domains. However, the large search space of training a neural network requires a significant amount of data, which makes the current RL algorithms not sample efficient. Motivated by the fact that many environments with continuous state space have smooth transitions, we propose to learn a smooth policy that behaves smoothly with respect to states. In contrast to policies parameterized by linear/reproducing kernel functions, where simple regularization techniques suffice to control smoothness, for neural network based reinforcement learning algorithms, there is no readily available solution to learn a smooth policy. In this paper, we develop a new training framework --- $\textbf{S}$mooth $\textbf{R}$egularized $\textbf{R}$einforcement $\textbf{L}$earning ($\textbf{SR}^2\textbf{L}$), where the policy is trained with smoothness-inducing regularization. Such regularization effectively constrains the search space of the learning algorithms and enforces smoothness in the learned policy. We apply the proposed framework to both on-policy (TRPO) and off-policy algorithm (DDPG). Through extensive experiments, we demonstrate that our method achieves improved sample efficiency.