136879 research outputs found
Sort by
Influence of Irrigation Water Quality and Quantity on Peanut Production in the Texas High Plains
Get PDF
Resource Allocation Optimization for Agriculture and Machine Learning Hardware Computing
Get PDFResource allocation has been a vital methodology throughout human history, enabling the pursuit of quality results by maintaining a delicate equilibrium among available resources. In this dissertation, I focus on developing efficient resource allocation strategies from agriculture to machine learning algorithms.
As the world population continues growing, the food supply has become a serious problem that needs to be tread carefully. Crops play a major role in the food supply, and as crop production becomes more mechanized and complex, an efficient strategy for making irrigation and chemigation control decisions towards a better crop yield and net return is extremely important. Due to the sensors data becoming more accessible, the limitations of production managers to effectively use the data for higher crop yield becomes more apparent. In this work, a double-deep Q-learning-based technique for irrigation and chemigation control has been evaluated. The proposed method is designed to achieve the maximum net return at harvest by automatically making irrigation and chemigation decisions during the growing season. By this approach, the proposed method is able to select the optimal or near-optimal irrigation and chemigation schedule toward a better net return.
As the implementation of deep learning algorithms continues to expand, the increasing complexity of Convolutional Neural Networks (CNNs) greatly underscores the demand for enhanced hardware acceleration. The need for such an acceleration raises a huge design challenge. The combined solution space for hardware design and CNN dataflow mapping is enormously large and it is discrete and lacks a well-behaved structure. The majority of previous works either focus on stochastic metaheuristics, such as genetic algorithm, which are typically slow for large-scale problems or heavily rely on expensive sampling, e.g., Gumbel Softmax based differentiable optimization and Bayesian optimization. In this work, an analytical model is proposed to evaluate the power and performance of CNN hardware design and dataflow solutions. Meanwhile, we also introduce a two-stage co-optimization method based on the model. The optimization method consists of nonlinear programming and parallel local search. One of the major contributions of this work is its matrix format based model, which enables the use of deep learning toolkits for efficient evaluation of power, performance values, and gradients during optimization stages. Compared with a weighted sum, the proposed method can provide better solutions for balancing the power-performance trade-off.
Rather than solely concentrating on dense CNNs, many real-world CNNs display sparsity. This characteristic has primarily been utilized in manual design processes and has received little attention in existing automatic optimization techniques. This dissertation presents the first systematic investigation on automatic dataflow and hardware optimization for sparse CNN computation, to the best of our knowledge. A differentiable PPA (Power Performance Area) model is developed to enable fast nonlinear optimization solving and massively parallel local search-based discretization. Experimental results on public domain testcases demonstrate the efficacy of the proposed approach
Black Fire authors
No full textPWAM, 1960-2022, Black Fire Authors, Author Page Count, as a percentage, zoom.Referenced in Chapter 4 of the book "Digital Literary Redlining: African American Anthologies, Digital Humanities, and the Canon.
Identification of Amino Acid Transporters and Their Regulators Involved in Wheat Grain Filling
No full textWheat (Triticum aestivum) is an important stable crops in the world. Wheat flour usage is dictated by its protein content and qualities, which is also associated with grain nitrogen content. To provide food for the increasing global population, nitrogen fertilizer used been increasing steadily for the past 100 years. Therefore, increasing the nitrogen use efficiency of wheat is of high interest to obtain better quality and higher yield. Recent studies have shown that by manipulating amino acid transporters, protein content and yield can be increased in both legume (i.e., pea) and grass (i.e., rice). I thus aim to explore this opportunity of utilizing these transporters in wheat to increase yield and protein content.
Firstly, bidirectional amino acid transporters, UMAMITs were identified in wheat via transcriptome and manual curation, and their expression are shown to be highly regulated by tissue type and development time. Given their redundancy, single transporter manipulation might not cause a measurable impact on phenotype. Thus, following the hypothesis that multiple transporters are regulated by a smaller number of hub genes, two complementary cultivars in protein content, TAM112 and TAM111 were selected to identify regulators that impact grain protein content. Via network analysis based on the transcriptome obtained from grains from these genotypes, several transcription factors were identified that might be central to their respective phenotypes. Lastly, the candidate transcription factors were further analyzed. The DAP-seq analysis has revealed NAC as a potential regulator of transporter regulator genes, and that all TFs tested were localized to the nucleus. For the future, several sgRNA vectors were generated to knockout these TFs to test their direct phenotypic impact.
A current bottleneck in following nitrogen in plants is the lack of high-resolution method that is non-destructive. Raman microscopy was examined as such a tool in measuring the content of nitrate for a small volume of live tissue. Using Arabidopsis root as target, it is shown that there is a concentration gradient of nitrate along the root axis that has never been reported. The results showed that Raman microscopy as a viable method to track nitrogen content from several cells
