QUANTUM LONG SHORT-TERM MEMORY
Explore the source record for details and available documents.
Engineering topics
Publications and source records attributed to Yoo, S.
Explore the source record for details and available documents.
High data rate detectors play an integral part in scientific research and their development is actively pursued at High Energy Physics (HEP) facilities around the world. Edge Machine Learning (ML) offers the ability to reduce data rates by integrating ML algorithms into Application Specific Integrated Circuits (ASICs) on the front end electronics. In this work, we explore a set of neural network architectures for predicting the peak amplitudes in the detector's sensor response. We have designed and synthesized several MLP based neural networks comparing their inference accuracy, power consumption, and area targeting for minimal latency. The neural networks are synthesized in a commercial 65nm process. The effect of quantizing the network's weights and biases on hardware performance and area is reported. We also conduct design space exploration to compare between design alternatives
Explore the source record for details and available documents.
Purpose: To enable near-real-time (<20sec) and interactive planning without compromising quality for whole breast RT treatment planning using tangential fields. Methods: Whole breast RT plans from 20 patients treated with single energy (SE, 6MV, 10 patients) or mixed energy (ME, 6/15MV, 10 patients) were randomly selected for model training. Additional 20 cases were used as validation cohort. The planning process for a new case consists of three fully automated steps:1. Energy Selection. A classification model automatically selects energy level. To build the energy selection model, principle component analysis (PCA) was applied to the digital reconstructed radiographs (DRRs) of training cases to extract anatomy-energy relationship.2. Fluence Estimation. Once energy is selected, a random forest (RF) model generates the initial fluence. This model summarizes the relationship between patient anatomy’s shape based features and the output fluence. 3. Fluence Fine-tuning. This step balances the overall dose contribution throughout the whole breast tissue by automatically selecting reference points and applying centrality correction. Fine-tuning works at beamlet-level until the dose distribution meets clinical objectives. Prior to finalization, physicians can also make patient-specific trade-offs between target coverage and high-dose volumes.The proposed method was validated by comparing auto-plans with manually generated clinical-plans using Wilcoxon Signed-Rank test. Results: In 19/20 cases the model suggested the same energy combination as clinical-plans. The target volume coverage V100% was 78.1±4.7% for auto-plans, and 79.3±4.8% for clinical-plans (p=0.12). Volumes receiving 105% Rx were 69.2±78.0cc for auto-plans compared to 83.9±87.2cc for clinical-plans (p=0.13). The mean V10Gy, V20Gy of the ipsilateral lung was 24.4±6.7%, 18.6±6.0% for auto plans and 24.6±6.7%, 18.9±6.1% for clinical-plans (p=0.04, <0.001). Total computational time for auto-plans was < 20s. Conclusion: We developed an automated method that generates breast radiotherapy plans with accurate energy selection, similar target volume coverage, reduced hotspot volumes, and significant reduction in planning time, allowing for near-real-time planning.
Email overload, even after spam filtering, presents a serious productivity challenge for busy professionals and executives. One solution is automated prioritization of incoming emails to ensure the most important are read and processed quickly, while others are processed later as/if time permits in declining priority levels. This paper presents a study of machine learning approaches to email prioritization into discrete levels, comparing ordinal regression versus classier cascades. Given the ordinal nature of discrete email priority levels, SVM ordinal regression would be expected to perform well, but surprisingly a cascade of SVM classifiers significantly outperforms ordinal regression for email prioritization. In contrast, SVM regression performs well -- better than classifiers -- on selected UCI data sets. This unexpected performance inversion is analyzed and results are presented, providing core functionality for email prioritization systems.