Computer Science

The choice of a proper learning rate is paramount for good Artificial Neural Network training and performance. In the past, one had to rely on experience and trial-and-error to find an adequate learning rate. Presently, a plethora of state of the art automatic methods exist that make the search for a good learning rate easier. While these techniques are effective and have yielded good results over the years, they are general solutions. This means the optimization of learning rate for specific network topologies remains largely unexplored. This work presents AutoLR, a framework that evolves Learning Rate Schedulers for a specific Neural Network Architecture using Structured Grammatical Evolution. The system was used to evolve learning rate policies that were compared with a commonly used baseline value for learning rate. Results show that training performed using certain evolved policies is more efficient than the established baseline and suggest that this approach is a viable means of improving a neural network's performance.

State-of-the-art Neural Network Architectures (NNAs) are challenging to design and implement efficiently in hardware. In the past couple of years, this has led to an explosion in research and development of automatic Neural Architecture Search (NAS) tools. AutomML tools are now used to achieve state of the art NNA designs and attempt to optimize for hardware usage and design. Much of the recent research in the auto-design of NNAs has focused on convolution networks and image recognition, ignoring the fact that a significant part of the workload in data centers is general-purpose deep neural networks. In this work, we develop and test a general multilayer perceptron (MLP) flow that can take arbitrary datasets as input and automatically produce optimized NNAs and hardware designs. We test the flow on six benchmarks. Our results show we exceed the performance of currently published MLP accuracy results and are competitive with non-MLP based results. We compare general and common GPU architectures with our scalable FPGA design and show we can achieve higher efficiency and higher throughput (outputs per second) for the majority of datasets. Further insights into the design space for both accurate networks and high performing hardware shows the power of co-design by correlating accuracy versus throughput, network size versus accuracy, and scaling to high-performance devices.

We demonstrate how an evolutionary algorithm can be extended with a curriculum learning process that selects automatically the environmental conditions in which the evolving agents are evaluated. The environmental conditions are selected so to adjust the level of difficulty to the ability level of the current evolving agents and so to challenge the weaknesses of the evolving agents. The method does not require domain knowledge and does not introduce additional hyperparameters. The results collected on two benchmark problems, that require to solve a task in significantly varying environmental conditions, demonstrate that the method proposed outperforms conventional algorithms and generates solutions that are robust to variations

A preference based multi-objective evolutionary algorithm is proposed for generating solutions in an automatically detected knee point region. It is named Automatic Preference based DI-MOEA (AP-DI-MOEA) where DI-MOEA stands for Diversity-Indicator based Multi-Objective Evolutionary Algorithm). AP-DI-MOEA has two main characteristics: firstly, it generates the preference region automatically during the optimization; secondly, it concentrates the solution set in this preference region. Moreover, the real-world vehicle fleet maintenance scheduling optimization (VFMSO) problem is formulated, and a customized multi-objective evolutionary algorithm (MOEA) is proposed to optimize maintenance schedules of vehicle fleets based on the predicted failure distribution of the components of cars. Furthermore, the customized MOEA for VFMSO is combined with AP-DI-MOEA to find maintenance schedules in the automatically generated preference region. Experimental results on multi-objective benchmark problems and our three-objective real-world application problems show that the newly proposed algorithm can generate the preference region accurately and that it can obtain better solutions in the preference region. Especially, in many cases, under the same budget, the Pareto optimal solutions obtained by AP-DI-MOEA dominate solutions obtained by MOEAs that pursue the entire Pareto front.

In this article, we apply the Free-Energy Principle to the question of motor primitives learning. An echo-state network is used to generate motor trajectories. We combine this network with a perception module and a controller that can influence its dynamics. This new compound network permits the autonomous learning of a repertoire of motor trajectories. To evaluate the repertoires built with our method, we exploit them in a handwriting task where primitives are chained to produce long-range sequences.

The human brain works in an unsupervised way, and more than one brain region is essential for lighting up intelligence. Inspired by this, we propose a brain-like heterogeneous network (BHN), which can cooperatively learn a lot of distributed representations and one global attention representation. By optimizing distributed, self-supervised, and gradient-isolated objective functions in a minimax fashion, our model improves its representations, which are generated from patches of pictures or frames of videos in experiments.

We recently proposed the S4NN algorithm, essentially an adaptation of backpropagation to multilayer spiking neural networks that use simple non-leaky integrate-and-fire neurons and a form of temporal coding known as time-to-first-spike coding. With this coding scheme, neurons fire at most once per stimulus, but the firing order carries information. Here, we introduce BS4NN, a modification of S4NN in which the synaptic weights are constrained to be binary (+1 or -1), in order to decrease memory and computation footprints. This was done using two sets of weights: firstly, real-valued weights, updated by gradient descent, and used in the backward pass of backpropagation, and secondly, their signs, used in the forward pass. Similar strategies have been used to train (non-spiking) binarized neural networks. The main difference is that BS4NN operates in the time domain: spikes are propagated sequentially, and different neurons may reach their threshold at different times, which increases computational power. We validated BS4NN on two popular benchmarks, MNIST and Fashion MNIST, and obtained state-of-the-art accuracies for this sort of networks (97.0% and 87.3% respectively) with a negligible accuracy drop with respect to real-valued weights (0.4% and 0.7%, respectively). We also demonstrated that BS4NN outperforms a simple BNN with the same architectures on those two datasets (by 0.2% and 0.9% respectively), presumably because it leverages the temporal dimension.

Balancing common disease treatment and epidemic control is a key objective of medical supplies procurement in hospitals during a pandemic such as COVID-19. This problem can be formulated as a bi-objective optimization problem for simultaneously optimizing the effects of common disease treatment and epidemic control. However, due to the large number of supplies, difficulties in evaluating the effects, and the strict budget constraint, it is difficult for existing evolutionary multiobjective algorithms to efficiently approximate the Pareto front of the problem. In this paper, we present an approach that first transforms the original high-dimensional, constrained multiobjective optimization problem to a low-dimensional, unconstrained multiobjective optimization problem, and then evaluates each solution to the transformed problem by solving a set of simple single-objective optimization subproblems, such that the problem can be efficiently solved by existing evolutionary multiobjective algorithms. We applied the transform-and-divide evolutionary optimization approach to six hospitals in Zhejiang Province, China, during the peak of COVID-19. Results showed that the proposed approach exhibits significantly better performance than that of directly solving the original problem. Our study has also shown that transform-and-divide evolutionary optimization based on problem-specific knowledge can be an efficient solution approach to many other complex problems and, therefore, enlarge the application field of evolutionary algorithms.

Introduction. Reservoir computing is a growing paradigm for simplified training of recurrent neural networks, with a high potential for hardware implementations. Numerous experiments in optics and electronics yield comparable performance to digital state-of-the-art algorithms. Many of the most recent works in the field focus on large-scale photonic systems, with tens of thousands of physical nodes and arbitrary interconnections. While this trend significantly expands the potential applications of photonic reservoir computing, it also complicates the optimisation of the high number of hyper-parameters of the system. Methods. In this work, we propose the use of Bayesian optimisation for efficient exploration of the hyper-parameter space in a minimum number of iteration. Results. We test this approach on a previously reported large-scale experimental system, compare it to the commonly used grid search, and report notable improvements in performance and the number of experimental iterations required to optimise the hyper-parameters. Conclusion. Bayesian optimisation thus has the potential to become the standard method for tuning the hyper-parameters in photonic reservoir computing.

Customisation in food properties is a challenging task involving optimisation of the production process with the demand to support computational creativity which is geared towards ensuring the presence of alternatives. This paper addresses the personalisation of beer properties in the specific case of craft beers where the production process is more flexible. We investigate the problem by using three swarm intelligence and evolutionary computation techniques that enable brewers to map physico-chemical properties to target organoleptic properties to design a specific brew. While there are several tools, using the original mathematical and chemistry formulas, or machine learning models that deal with the process of determining beer properties based on the pre-determined quantities of ingredients, the next step is to investigate an automated quantitative ingredient selection approach. The process is illustrated by a number of experiments designing craft beers where the results are investigated by "cloning" popular commercial brands based on their known properties. Algorithms performance is evaluated using accuracy, efficiency, reliability, population-diversity, iteration-based improvements and solution diversity. The proposed approach allows for the discovery of new recipes, personalisation and alternative high-fidelity reproduction of existing ones.