BACKGROUND: In plant breeding, many studies currently investigate the application of machine learning (ML) to genomic prediction, hoping for an improvement in prediction accuracy compared to standard models like Genomic Best Linear Unbiased Prediction (GBLUP). However, ML algorithms require much higher computational resources. This study aims to reduce the computational requirements and speed up training time by developing a novel autoencoder architecture inspired by haplotype blocks. Our approach incorporates prior knowledge on genetic linkage, inspired by haplotype block building, into the autoencoder architecture, resulting in a new encoded variable per haplotype block. We further modified our model into a semi-supervised version by adding available yield information. We used features extracted from the autoencoder’s block layer as inputs for Random Forest and GBLUP models to predict the yield of hybrid and inbred crops. RESULTS: Genomic prediction based on the extracted features maintained prediction accuracies equal to using the original marker data, even with a variable reduction of up to 98% and significantly reduced computation time. Prediction accuracy of the supervised component was in some cases equal to and in some lower than the prediction accuracy achieved using GBLUP. Effects estimated for haplotype block variants using our new method showed a high correlation to the blockwise sum of marker effects, which is the current standard approach for haplotype block effects. Correlation between the two block effect estimation approaches was very low for some blocks, which might indicate the incorporation of non-linear effects by the autoencoder. CONCLUSIONS: Our approach introduces a new perspective on processing haplotype blocks for genomic prediction, potentially providing more flexible modelling opportunities without the use of multiple binary dummy variables for each block variant. Additionally, training time for ML models may be significantly reduced by using the reduced feature sets generated using our method. By adding the semi-supervised component, the model is able to estimate values similar to marker effects for each block on yield. In future work, this may provide a new way of quantifying the importance of haplotype blocks for selection and breeding.

In communication systems, autoencoder refers to a system that replaces parts of the traditional transmitter and receiver of the baseband processing chain with artificial neural networks (ANNs). This allows to jointly train the system for an underlying channel model by reconstructing the input symbols at the output. Since the actual behavior of a real communication channel cannot be perfectly reproduced by an abstract model, it is necessary for the autoencoder to adapt to the changing conditions at runtime. Thus, online fine-tuning, in the form of ANN-retraining is of great importance. A platform able to satisfy the low-latency and low-power requirements of embedded communication systems are Field-programmable gate arrays (FPGAs). In this paper, we present an online-trainable low-power FPGA architecture for the receiver of an autoencoder-based communication chain. The architecture is embedded into an exploration framework that automatically determines the optimal degree of parallelism to minimize latency or power consumption. Our solutions achieve 2000×higher throughput than a high-performance GPU, draw 5×less power than an embedded CPU and are 5800×more energy efficient compared to an embedded GPU, for a batch size of one. To the best of our knowledge, this is the first FPGA-based autoencoder implementation for communication systems.

Spatially coupled serially concatenated codes (SC-SCCs) are constructed by coupling several classical turbo-like component codes. The resulting spatially coupled codes provide a close-to-capacity performance and low error floor, which have attracted a lot of interest in the past few years. The aim of this paper is to perform a comprehensive design space exploration to reveal different aspects of SC-SCCs, which is missing in the literature. More specifically, we investigate the effect of block length, coupling memory, decoding window size, and number of iterations on the decoding performance, complexity, latency, and throughput of SC-SCCs. To this end, we propose two decoding algorithms for the SC-SCCs:block-wiseandwindow-wisedecoders. For these, we present VLSI architectural templates and explore them based on building blocks implemented in 12nm FinFET technology. Linking architectural templates with the new algorithms, we demonstrate various tradeoffs between throughput, silicon area, latency, and decoding performance.

Recent work discovered that fitness landscapes induced by Iterated Local Search (ILS) may consist of multiple clusters, denoted as funnels or communities of local optima. Such studies exist only for perturbation operators (kicks) with low strength. We examine how different strengths of the ILS perturbation operator affect the number and size of clusters. We present an empirical study based on local optima networks from NK fitness landscapes. Our results show that a properly selected perturbation strength can help overcome the effect of ILS getting trapped in clusters of local optima. This has implications for designing effective ILS approaches in practice, where traditionally only small perturbations or complete restarts are applied, with the middle ground of intermediate perturbation strengths largely unexplored.