Stock trading has always been a challenging task due to the highly volatile nature of the stock market. Making trading decisions to generate profit is particularly difficult under such conditions. To address this, we propose four novel loss functions to drive trading strategies for a portfolio of stocks. These loss functions account for the potential profits or losses associated with buying or shorting respective stocks, enabling potentially any artificial neural network to directly learn trading decisions. Despite the high volatility in stock market fluctuations over time, training time-series models such as transformers using these loss functions resulted in trading strategies that generated strong returns on a portfolio of 50 different S&P 500 company stocks compared to benchmark reinforcement learning techniques and a baseline buy and hold method. As an example, using 2021, 2022 and 2023 as three test periods, the Crossformer model adapted with our best loss function was most consistent, resulting in returns of 51.42%, 51.04% and 48.62% respectively. By comparison, leading reinforcement learning models such as PPO, DDPG, and deep RL portfolio management techniques achieved lower returns for the same periods. Our approach was also shown to generalize across portfolios, achieving favorable performance on two other distinct portfolios with 30 and 10 companies. These results suggest that profit-guided loss functions provide a promising approach for aligning model training with downstream trading objectives.

Asynchronous parallel evolutionary algorithms improve computational efficiency but introduce evaluation-time bias: architectures with shorter evaluation times accumulate disproportionate reproductive opportunity, steering the search toward compact but potentially suboptimal solutions. While correction strategies such as Harada’s frequency-based parent selection and SWEET have been proposed, their behavior in island-based neuroevolution systems with periodic repopulation remains uncharacterized. We present a systematic ablation study comparing these bias correction strategies within EXAMM, a distributed asynchronous neuroevolution framework, across six island repopulation frequencies on three real-world time-series datasets: aviation flight recorder data, coal power plant sensor data and wind turbine sensor data. Each of the 24 experimental conditions is evaluated across 10 independent trials with a fixed budget of 10,000 genome evaluations, with statistical significance assessed via the Mann-Whitney U test against a no-repopulation vanilla baseline. Results show that repopulation frequency provides the strongest effect across the evaluated strategies.

This study investigates the effect of alternating growth and pruning phases on the evolution of neural architectures using the Evolutionary eXploration of Augmenting Memory Models (EXAMM). Our objective was to determine if this structured evolutionary process could guide the search towards more compact models without sacrificing predictive performance. We compared EXAMM-evolved architectures having traditional and five modern memory cells (simple, UGRNN, MGU, GRU, Delta-RNN, and LSTM) on two distinct time-series datasets: aviation flight recorder data and wind turbine sensor data. The alternating phases were controlled by enabling growth-promoting or growth-reducing mutation operations based on trigger frequencies of 50, 100, or 200 generated genomes, resulting in a 3 × 3 set of nine unique grow-shrink configurations. Results show that while there was no statistical difference in terms of model performance to a baseline without phases, the experiments with higher pruning to growth ratios had statistically significant decreases in model size, making this a simple but effective method for controlling network complexity.

Stock return forecasting is a critical application of time series forecasting in finance, facilitating informed trading and management decisions that can lead to substantial returns. However, for large investment portfolios, designing and fine-tuning models for individual stock predictions is time-consuming and computationally intensive. In this work, we propose using the neuroevolution-based neural architecture search algorithm, Evolutionary eXploration of Augmenting Memory Models (EXAMM), to evolve recurrent neural networks (RNNs) for stock return prediction. We compare the prediction performance of these evolved RNNs with that of state-of-the-art attention-based Transformer and deep learning models. Our results indicate that EXAMM-evolved RNNs outperform or achieve comparable performance with these models across 50 multivariate stock datasets and a combined high-dimensional dataset with 300 input features and 50 outputs. Additionally, they require orders of magnitude fewer parameters and can be evolved and operate efficiently using a minimal 8-core CPU configuration as opposed to expensive GPUs.

Evolutionary algorithms have shown substantial progress in recent years, especially in neural architecture search applications, or neuroevolution. Despite their effectiveness, analyzing and understanding the evolutionary paths these algorithms traverse to reach solutions remains challenging. Often these algorithms involve distributed computing strategies, which can include subpopulations or islands, and they explore massive or even unbounded search spaces which can include both weights and architecture, in both continuous and non-continuous domains. Manually examining individual solutions to understand the evolutionary dynamics is often infeasible due to large population sizes, large genome sizes, and high generation counts. This work introduces a new methodology for visualizing neuroevolution population dynamics called genetic distance projections, along with a novel neural network based method for generating these representations. We evaluate this methodology empirically and find it performs better than other traditional methods in generating these representations. We further validate the usefulness of these visualizations using case studies from EXAMM, a long standing neuroevolution algorithm, in which one case study even led to finding and fixing a bug in EXAMM's algorithm.

Parameter prediction is essential for many applications, facilitating insightful interpretation and decision-making. However, in many real life domains, such as power systems, medicine, and engineering, it can be very expensive to acquire ground truth labels for certain datasets as they may require extensive and expensive laboratory testing. In this work, we introduce a semi-supervised learning approach based on topological projections in self-organizing maps (SOMs), which significantly reduces the required number of labeled data points to perform parameter prediction, effectively exploiting information contained in large unlabeled datasets. While few-shot learning has seen significant advances in recent years, the majority of existing approaches focus on classification tasks, making our regression-based method particularly novel for continuous parameter estimation problems. Our proposed method first trains SOMs on unlabeled data followed by only using a minimal number of available labeled data points to assign targets to key best matching units (BMU). The values estimated for newly-encountered data points are computed utilizing the average of the N closest labeled data points in the SOM's U-matrix in tandem with a topological shortest path distance calculation scheme. The effectiveness of our approach has been validated through practical application in power engineering, specifically for estimating critical coal property values in coal power plants, where traditional laboratory testing is both time-consuming and costly. Our results indicate that the proposed minimally supervised model significantly outperforms traditional regression techniques, including linear and polynomial regression, Gaussian process regression, K-nearest neighbors, as well as deep neural network models and related clustering schemes.

This work introduces a novel, nature-inspired neural architecture search (NAS) algorithm based on ant colony optimization, Continuous Ant-based Neural Topology Search (CANTS), which utilizes synthetic ants that move over a continuous search space based on the density and distribution of pheromones, strongly inspired by how ants move in the real world. The paths taken by the ant agents through the search space are utilized to construct artificial neural networks (ANNs). This continuous search space allows CANTS to automate the design of ANNs of any size, removing a key limitation inherent to many current NAS algorithms that must operate within structures of a size predetermined by the user. CANTS employs a distributed asynchronous strategy which allows it to scale to large-scale high performance computing resources, works with a variety of recurrent memory cell structures, and uses of a communal weight sharing strategy to reduce training time. The proposed procedure is evaluated on three real-world, time series prediction problems in the field of power systems and compared to two state-of-the-art algorithms. Results show that CANTS is able to provide improved or competitive results on all of these problems while also being easier to use, requiring half the number of user-specified hyper-parameters.

Identifying phases of flight is important in the field of general aviation, as knowing which phase of flight data is collected from aircraft flight data recorders can aid in the more effective detection of safety or hazardous events. General aviation flight data for phase of flight identification is usually per-second data, comes on a large scale, and is class imbalanced. It is expensive to manually label the data and training classification models usually faces class imbalance problems. This work investigates the use of a novel method for minimally supervised self-organizing maps (MS-SOMs) which utilize nearest neighbor majority votes in the SOM U-matrix for class estimation. Results show that the proposed method can reach or exceed a naive SOM approach which utilized a full data file of labeled data, with only 30 labeled datapoints per class. Additionally, the minimally supervised SOM is significantly more robust to the class imbalance of the phase of flight data. These results highlight how little data is required for effective phase of flight identification.

Stock return forecasting is a major component of numerous finance applications. Predicted stock returns can be incorporated into portfolio trading algorithms to make informed buy or sell decisions which can optimize returns. In such portfolio trading applications, the predictive performance of a time series forecasting model is crucial. In this work, we propose the use of the Evolutionary eXploration of Augmenting Memory Models (EXAMM) algorithm to progressively evolve recurrent neural networks (RNNs) for stock return predictions. RNNs are evolved independently for each stocks and portfolio trading decisions are made based on the predicted stock returns. The portfolio used for testing consists of the 30 companies in the Dow-Jones Index (DJI) with each stock have the same weight. Results show that using these evolved RNNs and a simple daily long-short strategy can generate higher returns than both the DJI index and the S&P 500 Index for both 2022 (bear market) and 2023 (bull market).

Multi-armed bandits and contextual multi-armed bandits have demonstrated their proficiency in a variety of application areas. However, these models are highly susceptible to volatility and often exhibit knowledge gaps due to a limited understanding of future states. In this paper, we propose a new bandit framework for what we refer to as informed contextual multi armed bandits (iCMABs) to mitigate these gaps, facilitating "informed" decisions based on predicted future contexts. The performance of an iCMAB is thus highly dependent on the accuracy of the forecast it uses. We examine the use of recurrent neural networks (RNNs) evolved through the EXAMM neuroevolution algorithm as compared to other time series forecasting (TSF) methods and evaluate our iCMAB framework's ability to make stock market trading decisions for the Dow-Jones Index (DJI) in comparison to other decision making strategies using these forecasts. Our results demonstrate that an iCMAB, driven by evolved RNN architectures, performs better than statistical TSF methods, fixed architecture RNNs for TSF, and other CMAB methods. Using evolved RNNs, iCMAB is able to achieve the highest return of over 21%, a similar to 7% improvement over not incorporating forecasted values, and a similar to 5% improvement over DJI's return for that time period.

Designing machine learning models involves determining not only the network architecture, but also non-architectural elements such as training hyperparameters. Further confounding this problem, different architectures and datasets will perform more optimally with different hyperparameters. This problem is exacerbated for neuroevolution (NE) and neural architecture search (NAS) algorithms, which can generate and train architectures with a wide variety of architectures in order to find optimal architectures. In such algorithms, if hyperparameters are fixed, then suboptimal architectures can be found as they will be biased towards the fixed parameters. This paper evaluates the use of the simplex hyperparameter optimization (SHO) method, which allows co-evolution of hyperparameters over the course of a NE algorithm, allowing the NE algorithm to simultaneously optimize both network architectures and hyperparameters. SHO has been previously shown to be able to optimize hyperparameters for convolutional neural networks using traditional stochastic gradient descent with Nesterov momentum, and this work extends on this to evaluate SHO for evolving recurrent neural networks with additional modern weight optimizers such as RMSProp and Adam. Results show that incorporating SHO into the neuroevolution process not only enables finding better performing architectures but also faster convergence to optimal architectures across all datasets and optimization methods tested.

Time series forecasting (TSF) is one of the most important tasks in data science given the fact that accurate time series (TS) predictive models play a major role across a wide variety of domains including finance, transportation, health care, and power systems. Real-world utilization of machine learning (ML) typically involves (pre-)training models on collected, historical data and then applying them to unseen data points. However, in real-world applications, time series data streams are usually non-stationary and trained ML models usually, over time, face the problem of data or concept drift. To address this issue, models must be periodically retrained or redesigned, which takes significant human and computational resources. Additionally, historical data may not even exist to re-train or re-design model with. As a result, it is highly desirable that models are designed and trained in an online fashion. This work presents the Online NeuroEvolution-based Neural Architecture Search (ONE-NAS) algorithm, which is a novel neural architecture search method capable of automatically designing and dynamically training recurrent neural networks (RNNs) for online forecasting tasks. Without any pre-training, ONE-NAS utilizes populations of RNNs that are continuously updated with new network structures and weights in response to new multivariate input data. ONE-NAS is tested on real-world, large-scale multivariate wind turbine data as well as the univariate Dow Jones Industrial Average (DJIA) dataset. Results demonstrate that ONE-NAS outperforms traditional statistical time series forecasting methods, including online linear regression, fixed long short-term memory (LSTM) and gated recurrent unit (GRU) models trained online, as well as state-of-the-art, online ARIMA strategies.

Continuous Ant-based Topology Search (CANTS) is a previously introduced novel nature-inspired neural architecture search (NAS) algorithm that is based on ant colony optimization (ACO). CANTS utilizes a continuous search space to indirectly-encode a neural architecture search space. Synthetic ant agents explore CANTS’ continuous search space based on the density and distribution of pheromones, strongly inspired by how ants move in the real world. This continuous search space allows CANTS to automate the design of artificial neural networks (ANNs) of any size, removing a key limitation inherent to many current NAS algorithms that must operate within structures of a size that is predetermined by the user. This work expands CANTS by adding a fourth dimension to its search space representing potential neural synaptic weights. Adding this extra dimension allows CANTS agents to optimize both the architecture as well as the weights of an ANN without applying backpropagation (BP), which leads to a significant reduction in the time consumed in the optimization process: at least an average of 96% less time consumption with very competitive optimization performance, if not better. The experiments of this study – using real-world data – demonstrate that the BP-Free CANTS algorithm exhibits highly competitive performance compared to both CANTS and ANTS while requiring significantly less operation time.

Recent studies have shown that the performance of evolutionary neural architecture search (i.e., neuroevolution) algorithms can be significantly improved by the use of island based strategies which periodically experience extinction and repopulation events. Further, it has been shown that the simplex hyperparameter optimization (SHO) method can also improve neuroevolution (NE) performance by optimizing neural network training hyperparameters while the NE algorithm also trains and designs neural networks. This work provides an extensive examination of combining island repopulation events with five different island-based variations of SHO. These methods are evaluated for the evolution of recurrent neural networks for the challenging problem of multivariate time series forecasting on two real world datasets. We show with statistical significance that adding repopulation to the SHO variants in almost every case improves performance, and for those that does there is no statistical difference. In addition, we find that one variant in particular, multi-island, random island best genome (MIRIB) performs the best across all experiment types.

Time series forecasting (TSF) is one of the most important tasks in data science, as accurate time series (TS) predictions can drive and advance a wide variety of domains including finance, transportation, health care, and power systems. However, real-world utilization of machine learning (ML) models for TSF suffers due to data drift. To address this, models must be periodically retained or redesigned, which requires significant human and computational resources. This work presents the Online NeuroEvolution based Neural Architecture Search (ONE-NAS) algorithm, which to the authors' knowledge is the first neural architecture search algorithm capable of automatically designing and training new recurrent neural networks (RNNs) in an online setting. Without any pretraining, ONE-NAS utilizes populations of RNNs which are continuously updated with new network structures and weights in response to new multivariate input data. ONE-NAS is tested on real-world large-scale multivariate wind turbine data and is shown to outperform traditional statistical time series forecasting, including naive, moving average, and exponential smoothing methods.

Time series forecasting (TSF) is one of the most important tasks in data science, as accurate time series (TS) predictions can drive and advance a wide variety of domains including finance, transportation, health care, and power systems. However, real-world utilization of machine learning (ML) models for TSF suffers due to data drift. To address this, models must be periodically retained or redesigned, which requires significant human and computational resources. This work presents the Online NeuroEvolution based Neural Architecture Search (ONE-NAS) algorithm, which to the authors' knowledge is the first neural architecture search algorithm capable of automatically designing and training new recurrent neural networks (RNNs) in an online setting. Without any pretraining, ONE-NAS utilizes populations of RNNs which are continuously updated with new network structures and weights in response to new multivariate input data. ONE-NAS is tested on real-world large-scale multivariate wind turbine data and is shown to outperform traditional statistical time series forecasting, including naive, moving average, and exponential smoothing methods.

Weight initialization is critical in being able to successfully train artificial neural networks (ANNs), and even more so for recurrent neural networks (RNNs) which can easily suffer from vanishing and exploding gradients. In neuroevolution, where evolutionary algorithms are applied to neural architecture search, weights typically need to be initialized at three different times: when the initial genomes (ANN architectures) are created, when offspring genomes are generated by crossover, and when new nodes or edges are created during mutation. This work explores the difference between the state-of-the-art Xavier and Kaiming methods, and novel Lamarckian weight inheritance for weight initialization during crossover and mutation operations. These are examined using the Evolutionary eXploration of Augmenting Memory Models (EXAMM) neuroevolution algorithm, which is capable of evolving RNNs with a variety of modern memory cells (e.g., LSTM, GRU, MGU, UGRNN and Delta-RNN cells) as well as recurrent connections with varying time skips through a high performance island based distributed evolutionary algorithm. Results show that with statistical significance, the Lamarckian strategy outperforms both Kaiming and Xavier weight initialization, can speed neuroevolution by requiring less backpropagation epochs to be evaluated per genome, and that the neuroevolutionary process provides further benefits to neural network weight optimization.

Neuroevolution commonly uses speciation strategies to better explore the search space of neural network architectures. One such speciation strategy is the use of islands, which are also popular in improving the performance of distributed evolutionary algorithms. However, islands may experience stagnation, which prevents their convergence towards better solutions and can result in wasted computation. This work evaluates utilizing an island extinction and repopulation mechanism to avoid premature convergence using Evolutionary eXploration of Augmenting Memory Models (EXAMM), an asynchronous island based neuroevolution algorithm that progressively evolves recurrent neural networks (RNNs). In island extinction and repopulation, all members of the worst performing island are erased periodically and repopulated with mutated versions of the global best RNN. This island based strategy is additionally compared to NEAT’s (NeuroEvolution of Augmenting Topologies) speciation strategy. Experiments were performed using two different real-world time series datasets (coal-fired power plant and aviation flight data). With statistical significance, results show that in addition to being more scalable, this island extinction and repopulation strategy evolves better global best genomes than both EXAMM’s original island based strategy and NEAT’s speciation strategy. The extinction and repopulation strategy is easy to implement, and can be generically applied to other neuroevolution algorithms.

This work presents how the Evolutionary eXploration of Augmenting Memory Models (EXAMM) neuroevolution algorithm is incorporated into Microbeam Technologies' condition-based monitoring power plant optimization software using a workflow that integrates coal-fired power plant data collection, evolved RNN predictions and analytic performance indices predictions. To the authors' knowledge, it is the first use of a neuroevolution strategy to evolve recurrent neural networks (RNNs) for forecasting of power plant parameters where the evolved networks have been incorporated into production software used at a coal-fired power plant. A preliminary exploration of the plant's performance shows that after incorporating this software, the amount of revenue lost due to power plant derates and outages decreased by $7.3 million, a savings of 42%, and increased efficiency under medium and low load conditions. A further investigation of the effect of training sequence length and time series data normalization methods on evolving and training RNNs for this system is given, providing practical results useful for real world time series forecasting. It is shown that dividing long time series sequences up into shortened training sequences can dramatically speed up training, and that using different normalization methods (min-max vs. z-score) can provide statistically significant results, dependent on the data sets.

Transfer learning entails taking an artificial neural network (ANN) that is trained on a source dataset and adapting it to a new target dataset. While this has been shown to be quite powerful, its use has generally been restricted by architectural constraints. Previously, in order to reuse and adapt an ANN's internal weights and structure, the underlying topology of the ANN being transferred across tasks must remain mostly the same while a new output layer is attached, discarding the old output layer's weights. This work introduces network-aware adaptive structure transfer learning (N-ASTL), an advancement over prior efforts to remove this restriction. N-ASTL utilizes statistical information related to the source network's topology and weight distribution in order to inform how new input and output neurons are to be integrated into the existing structure. Results show improvements over prior state-of-the-art, including the ability to transfer in challenging real-world datasets not previously possible and improved generalization over RNNs without transfer.