Deployment of deep learning in different fields and industries is growing day by day due to its performance, which relies on the availability of data and compute. Data is often crowd-sourced and contains sensitive information about its contributors, which leaks into models that are trained on it. To achieve rigorous privacy guarantees, differentially private training mechanisms are used. However, it has recently been shown that differential privacy can exacerbate existing biases in the data and have disparate impacts on the accuracy of different subgroups of data. In this paper, we aim to study these effects within differentially private deep learning. Specifically, we aim to study how different levels of imbalance in the data affect the accuracy and the fairness of the decisions made by the model, given different levels of privacy. We demonstrate that even small imbalances and loose privacy guarantees can cause disparate impacts.

Risk is 6 player game with significant randomness and a large game-tree complexity which poses a challenge to creating an agent to play the game effectively. Previous AIs focus on creating high-level handcrafted features determine agent decision making. In this project, I create D.A.D, A Risk agent using temporal difference reinforcement learning to train a Deep Neural Network including a Graph Convolutional Network to evaluate player positions. This is used in a game-tree to select optimal moves. This allows minimal handcrafting of knowledge into the AI, assuring input features are as low-level as possible to allow the network to extract useful and sophisticated features itself, even with the network starting from a random initialisation. I also tackle the issue of non-determinism in Risk by introducing a new method of interpreting attack moves necessary for the search. The result is an AI which wins 35% of the time versus 5 of best inbuilt AIs in Lux Delux, a Risk variant.

Device identification is the process of identifying a device on Internet without using its assigned network or other credentials. The sharp rise of usage in Internet of Things (IoT) devices has imposed new challenges in device identification due to a wide variety of devices, protocols and control interfaces. In a network, conventional IoT devices identify each other by utilizing IP or MAC addresses, which are prone to spoofing. Moreover, IoT devices are low power devices with minimal embedded security solution. To mitigate the issue in IoT devices, fingerprint (DFP) for device identification can be used. DFP identifies a device by using implicit identifiers, such as network traffic (or packets), radio signal, which a device used for its communication over the network. These identifiers are closely related to the device hardware and software features. In this paper, we exploit TCP/IP packet header features to create a device fingerprint utilizing device originated network packets. We present a set of three metrics which separate some features from a packet which contribute actively for device identification. To evaluate our approach, we used publicly accessible two datasets. We observed the accuracy of device genre classification 99.37% and 83.35% of accuracy in the identification of an individual device from IoT Sentinel dataset. However, using UNSW dataset device type identification accuracy reached up to 97.78%.

The ability to exploit prior experience to solve novel problems rapidly is a hallmark of biological learning systems and of great practical importance for artificial ones. In the meta reinforcement learning literature much recent work has focused on the problem of optimizing the learning process itself. In this paper we study a complementary approach which is conceptually simple, general, modular and built on top of recent improvements in off-policy learning. The framework is inspired by ideas from the probabilistic inference literature and combines robust off-policy learning with a behavior prior, or default behavior that constrains the space of solutions and serves as a bias for exploration; as well as a representation for the value function, both of which are easily learned from a number of training tasks in a multi-task scenario. Our approach achieves competitive adaptation performance on hold-out tasks compared to meta reinforcement learning baselines and can scale to complex sparse-reward scenarios.

We introduce Bootstrap Your Own Latent (BYOL), a new approach to self-supervised image representation learning. BYOL relies on two neural networks, referred to as online and target networks, that interact and learn from each other. From an augmented view of an image, we train the online network to predict the target network representation of the same image under a different augmented view. At the same time, we update the target network with a slow-moving average of the online network. While state-of-the art methods rely on negative pairs, BYOL achieves a new state of the art without them. BYOL reaches $74.3\%$ top-1 classification accuracy on ImageNet using a linear evaluation with a ResNet-50 architecture and $79.6\%$ with a larger ResNet. We show that BYOL performs on par or better than the current state of the art on both transfer and semi-supervised benchmarks. Our implementation and pretrained models are given on GitHub.

Artificial intelligence (AI) use in dermatology is primed to become a powerful tool in skin cancer assessment, but it remains to be seen how diagnostic devices in dermatology will influence decision making in the clinic and affect patient outcomes, according to the authors of a Perspective published online today by the Medical Journal of Australia. In dermatology the primary focus for the use of AI has been on developing machine learning systems that facilitate classification and decision support for skin cancer management. "Recent studies show that machine learning algorithms have the potential to surpass the diagnostic performance of experts, and the challenge now is how to implement this new technology safely into clinical practice," wrote the authors, led by Associate Professor Victoria Mar, a consultant dermatologist and Director of the Victorian Melanoma Service at Alfred Hospital. "There are two potentially negative implications for clinical practice: first, clinicians may have difficulty upskilling by following the algorithms' outputs; and second, there exists the potential for deskilling and underperforming due to an over-reliance on technology. Algorithm performance is dependent on both the size and quality of the training image dataset and on whether the algorithm is used in situations for which it was intended," wrote Mar and colleagues.

We present a general approach, based on an exponential inequality, to derive bounds on the generalization error of randomized learning algorithms. Using this approach, we provide bounds on the average generalization error as well as bounds on its tail probability, for both the PAC-Bayesian and single-draw scenarios. Specifically, for the case of subgaussian loss functions, we obtain novel bounds that depend on the information density between the training data and the output hypothesis. When suitably weakened, these bounds recover many of the information-theoretic available bounds in the literature. We also extend the proposed exponential-inequality approach to the setting recently introduced by Steinke and Zakynthinou (2020), where the learning algorithm depends on a randomly selected subset of the available training data. For this setup, we present bounds for bounded loss functions in terms of the conditional information density between the output hypothesis and the random variable determining the subset choice, given all training data. Through our approach, we recover the average generalization bound presented by Steinke and Zakynthinou (2020) and extend it to the PAC-Bayesian and single-draw scenarios. For the single-draw scenario, we also obtain novel bounds in terms of the conditional $\alpha$-mutual information and the conditional maximal leakage.

Given the recent successes of Deep Learning in AI there has been increased interest in the role and need for explanations in machine learned theories. A distinct notion in this context is that of Michie's definition of Ultra-Strong Machine Learning (USML). USML is demonstrated by a measurable increase in human performance of a task following provision to the human of a symbolic machine learned theory for task performance. A recent paper demonstrates the beneficial effect of a machine learned logic theory for a classification task, yet no existing work has examined the potential harmfulness of machine's involvement in human learning. This paper investigates the explanatory effects of a machine learned theory in the context of simple two person games and proposes a framework for identifying the harmfulness of machine explanations based on the Cognitive Science literature. The approach involves a cognitive window consisting of two quantifiable bounds and it is supported by empirical evidence collected from human trials. Our quantitative and qualitative results indicate that human learning aided by a symbolic machine learned theory which satisfies a cognitive window has achieved significantly higher performance than human self learning. Results also demonstrate that human learning aided by a symbolic machine learned theory that fails to satisfy this window leads to significantly worse performance than unaided human learning.

Currently, unmanned aerial vehicles, such as drones, are becoming a part of our lives and reaching out to many areas of society, including the industrialized world. A common alternative to control the movements and actions of the drone is through unwired tactile interfaces, for which different remote control devices can be found. However, control through such devices is not a natural, human-like communication interface, which sometimes is difficult to master for some users. In this work, we present a domain-based speech recognition architecture to effectively control an unmanned aerial vehicle such as a drone. The drone control is performed using a more natural, human-like way to communicate the instructions. Moreover, we implement an algorithm for command interpretation using both Spanish and English languages, as well as to control the movements of the drone in a simulated domestic environment. The conducted experiments involve participants giving voice commands to the drone in both languages in order to compare the effectiveness of each of them, considering the mother tongue of the participants in the experiment. Additionally, different levels of distortion have been applied to the voice commands in order to test the proposed approach when facing noisy input signals. The obtained results show that the unmanned aerial vehicle is capable of interpreting user voice instructions achieving an improvement in speech-to-action recognition for both languages when using phoneme matching in comparison to only using the cloud-based algorithm without domain-based instructions. Using raw audio inputs, the cloud-based approach achieves 74.81% and 97.04% accuracy for English and Spanish instructions respectively, whereas using our phoneme matching approach the results are improved achieving 93.33% and 100.00% accuracy for English and Spanish languages.

Narratives are fundamental to our perception of the world and are pervasive in all activities that involve the representation of events in time. Yet, modern online information systems do not incorporate narratives in their representation of events occurring over time. This article aims to bridge this gap, combining the theory of narrative representations with the data from modern online systems. We make three key contributions: a theory-driven computational representation of narratives, a novel extraction algorithm to obtain these representations from data, and an evaluation of our approach. In particular, given the effectiveness of visual metaphors, we employ a route map metaphor to design a narrative map representation. The narrative map representation illustrates the events and stories in the narrative as a series of landmarks and routes on the map. Each element of our representation is backed by a corresponding element from formal narrative theory, thus providing a solid theoretical background to our method. Our approach extracts the underlying graph structure of the narrative map using a novel optimization technique focused on maximizing coherence while respecting structural and coverage constraints. We showcase the effectiveness of our approach by performing a user evaluation to assess the quality of the representation, metaphor, and visualization. Evaluation results indicate that the Narrative Map representation is a powerful method to communicate complex narratives to individuals. Our findings have implications for intelligence analysts, computational journalists, and misinformation researchers.