Augusta Ada King, Countess of Lovelace, widely regarded as the world's first computer programmer, when talking about the Analytical Engine said, "The Analytical Engine has no pretensions whatever to originate anything" [1]. Hence, it is safe to say that the question "Can computers think?", in some form, not only predates the concept of Artificial Intelligence (AI) but is almost as old as the Analytical Engine. This question has stimulated the minds of pioneers and researchers from different domains including computer science, mathematics, psychology and philosophy. This essay delves into some of the important facets of this question. It is primarily driven by the thoughts and arguments of Alan M. Turing and John R. Searle, two pioneers who have extensively explored this question.

Deep learning models have become an increasingly preferred option for biometric recognition systems, such as speaker recognition. SincNet, a deep neural network architecture, gained popularity in speaker recognition tasks due to its parameterized sinc functions that allow it to work directly on the speech signal. The original SincNet architecture uses the softmax loss, which may not be the most suitable choice for recognition-based tasks. Such loss functions do not impose inter-class margins nor differentiate between easy and hard training samples. Curriculum learning, particularly those leveraging angular margin-based losses, has proven very successful in other biometric applications such as face recognition. The advantage of such a curriculum learning-based techniques is that it will impose inter-class margins as well as taking to account easy and hard samples. In this paper, we propose Curricular SincNet (CL-SincNet), an improved SincNet model where we use a curricular loss function to train the SincNet architecture. The proposed model is evaluated on multiple datasets using intra-dataset and inter-dataset evaluation protocols. In both settings, the model performs competitively with other previously published work. In the case of inter-dataset testing, it achieves the best overall results with a reduction of 4\% error rate compare to SincNet and other published work.

Convolutional Neural Networks (CNNs) are widely used in deep learning applications, e.g. visual systems, robotics etc. However, existing software solutions are not efficient. Therefore, many hardware accelerators have been proposed optimizing performance, power and resource utilization of the implementation. Amongst existing solutions, Field Programmable Gate Array (FPGA) based architecture provides better cost-energy-performance trade-offs as well as scalability and minimizing development time. In this paper, we present a model-independent reconfigurable co-processing architecture to accelerate CNNs. Our architecture consists of parallel Multiply and Accumulate (MAC) units with caching techniques and interconnection networks to exploit maximum data parallelism. In contrast to existing solutions, we introduce limited precision 32 bit Q-format fixed point quantization for arithmetic representations and operations. As a result, our architecture achieved significant reduction in resource utilization with competitive accuracy. Furthermore, we developed an assembly-type microinstructions to access the co-processing fabric to manage layer-wise parallelism, thereby making re-use of limited resources. Finally, we have tested our architecture up to 9x9 kernel size on Xilinx Virtex 7 FPGA, achieving a throughput of up to 226.2 GOp/S for 3x3 kernel size.

Federated learning (FL) is a distributed machine learning approach involving multiple clients collaboratively training a shared model. Such a system has the advantage of more training data from multiple clients, but data can be non-identically and independently distributed (non-i.i.d.). Privacy and integrity preserving features such as differential privacy (DP) and robust aggregation (RA) are commonly used in FL. In this work, we show that on common deep learning tasks, the performance of FL models differs amongst clients and situations, and FL models can sometimes perform worse than local models due to non-i.i.d. data. Secondly, we show that incorporating DP and RA degrades performance further. Then, we conduct an ablation study on the performance impact of different combinations of common personalization approaches for FL, such as finetuning, mixture-of-experts ensemble, multi-task learning, and knowledge distillation. It is observed that certain combinations of personalization approaches are more impactful in certain scenarios while others always improve performance, and combination approaches are better than individual ones. Most clients obtained better performance with combined personalized FL and recover from performance degradation caused by non-i.i.d. data, DP, and RA.

Even with generational improvements in DRAM technology, memory access latency still remains the major bottleneck for application accelerators, primarily due to limitations in memory interface IPs which cannot fully account for variations in target applications, the algorithms used, and accelerator architectures. Since developing memory controllers for different applications is time-consuming, this paper introduces a modular and programmable memory controller that can be configured for different target applications on available hardware resources. The proposed memory controller efficiently supports cache-line accesses along with bulk memory transfers. The user can configure the controller depending on the available logic resources on the FPGA, memory access pattern, and external memory specifications. The modular design supports various memory access optimization techniques including, request scheduling, internal caching, and direct memory access. These techniques contribute to reducing the overall latency while maintaining high sustained bandwidth. We implement the system on a state-of-the-art FPGA and evaluate its performance using two widely studied domains: graph analytics and deep learning workloads. We show improved overall memory access time up to 58% on CNN and GCN workloads compared with commercial memory controller IPs.

Activation functions play a pivotal role in determining the training dynamics and neural network performance. The widely adopted activation function ReLU despite being simple and effective has few disadvantages including the Dying ReLU problem. In order to tackle such problems, we propose a novel activation function called Serf which is self-regularized and nonmonotonic in nature. Like Mish, Serf also belongs to the Swish family of functions. Based on several experiments on computer vision (image classification and object detection) and natural language processing (machine translation, sentiment classification and multimodal entailment) tasks with different state-of-the-art architectures, it is observed that Serf vastly outperforms ReLU (baseline) and other activation functions including both Swish and Mish, with a markedly bigger margin on deeper architectures. Ablation studies further demonstrate that Serf based architectures perform better than those of Swish and Mish in varying scenarios, validating the effectiveness and compatibility of Serf with varying depth, complexity, optimizers, learning rates, batch sizes, initializers and dropout rates. Finally, we investigate the mathematical relation between Swish and Serf, thereby showing the impact of preconditioner function ingrained in the first derivative of Serf which provides a regularization effect making gradients smoother and optimization faster.

COVID-19 continues to cause a significant impact on public health. To minimize this impact, policy makers undertake containment measures that however, when carried out disproportionately to the actual threat, as a result if errorneous threat assessment, cause undesirable long-term socio-economic complications. In addition, macro-level or national level decision making fails to consider the localized sensitivities in small regions. Hence, the need arises for region-wise threat assessments that provide insights on the behaviour of COVID-19 through time, enabled through accurate forecasts. In this study, a forecasting solution is proposed, to predict daily new cases of COVID-19 in regions small enough where containment measures could be locally implemented, by targeting three main shortcomings that exist in literature; the unreliability of existing data caused by inconsistent testing patterns in smaller regions, weak deploy-ability of forecasting models towards predicting cases in previously unseen regions, and model training biases caused by the imbalanced nature of data in COVID-19 epi-curves. Hence, the contributions of this study are three-fold; an optimized smoothing technique to smoothen less deterministic epi-curves based on epidemiological dynamics of that region, a Long-Short-Term-Memory (LSTM) based forecasting model trained using data from select regions to create a representative and diverse training set that maximizes deploy-ability in regions with lack of historical data, and an adaptive loss function whilst training to mitigate the data imbalances seen in epi-curves. The proposed smoothing technique, the generalized training strategy and the adaptive loss function largely increased the overall accuracy of the forecast, which enables efficient containment measures at a more localized micro-level.

Crystal-structure phase mapping is a core, long-standing challenge in materials science that requires identifying crystal structures, or mixtures thereof, in synthesized materials. Materials science experts excel at solving simple systems but cannot solve complex systems, creating a major bottleneck in high-throughput materials discovery. Herein we show how to automate crystal-structure phase mapping. We formulate phase mapping as an unsupervised pattern demixing problem and describe how to solve it using Deep Reasoning Networks (DRNets). DRNets combine deep learning with constraint reasoning for incorporating scientific prior knowledge and consequently require only a modest amount of (unlabeled) data. DRNets compensate for the limited data by exploiting and magnifying the rich prior knowledge about the thermodynamic rules governing the mixtures of crystals with constraint reasoning seamlessly integrated into neural network optimization. DRNets are designed with an interpretable latent space for encoding prior-knowledge domain constraints and seamlessly integrate constraint reasoning into neural network optimization. DRNets surpass previous approaches on crystal-structure phase mapping, unraveling the Bi-Cu-V oxide phase diagram, and aiding the discovery of solar-fuels materials.

Epilepsy is a disease that affects around 50 million people worldwide, 3-5% of which had seizures triggered by pulsating lights, or by patterns with light and dark areas, a condition known as photosensitive epilepsy (PSE). Video content with fast luminance variations, or with spatial patterns of high contrast - referred to as epileptogenic visual content - may also induce seizures on viewers with PSE, and even cause discomfort on users not affected by this disease. To avoid this type of effects, harmful visual content should be detected before being distributed or displayed. The problem has amplified during the COVID-19 pandemic, as all education, recreational and social activities move online. Estimates state that about 4.56 billion people have access to the internet in 2021, a number that accounts for nearly 60 percent of the world's total population. Out of 60 percent, more than 9 out of 10 people consume video content on a daily basis. A study done by the American Academy of Children Surveys shows that children from the age of 8-12 have had a 25% increase in average screen times from 4.5 hours a day to about 6 hours of screen time everyday.

This paper describe an hybrid agent trained to play in Fantasy Football AI which participated in the Bot Bowl III competition. The agent, MimicBot, is implemented using a specifically designed deep policy network and trained using a combination of imitation and reinforcement learning. Previous attempts in using a reinforcement learning approach in such context failed for a number of reasons, e.g. due to the intrinsic randomness in the environment and the large and uneven number of actions available, with a curriculum learning approach failing to consistently beat a randomly paying agent. Currently no machine learning approach can beat a scripted bot which makes use of the domain knowledge on the game. Our solution, thanks to an imitation learning and a hybrid decision-making process, consistently beat such scripted agents. Moreover we shed lights on how to more efficiently train in a reinforcement learning setting while drastically increasing sample efficiency. MimicBot is the winner of the Bot Bowl III competition, and it is currently the state-of-the-art solution.