A5017932766- Energy Load and Power Forecasting
- Computational Drug Discovery Methods
- Time Series Analysis and Forecasting
- Cell Image Analysis Techniques
- Gene expression and cancer classification
- Neural Networks and Applications
- Machine Learning and Data Classification
- Machine Learning in Materials Science
- Sparse and Compressive Sensing Techniques
- Music and Audio Processing
- Image and Signal Denoising Methods
- Rough Sets and Fuzzy Logic
- Anomaly Detection Techniques and Applications
- Advanced Data Storage Technologies
- Data Mining Algorithms and Applications
- Medical Imaging Techniques and Applications
- Computer Graphics and Visualization Techniques
- Advanced Image Processing Techniques
- Bioinformatics and Genomic Networks
- Respiratory and Cough-Related Research
- Explainable Artificial Intelligence (XAI)
- Advanced Multi-Objective Optimization Algorithms
- Advanced biosensing and bioanalysis techniques
- Image Retrieval and Classification Techniques
- Machine Learning and Algorithms
Lawrence Livermore National Laboratory
2012-2023
Beihang University
2021-2023
Abstract To enable personalized cancer treatment, machine learning models have been developed to predict drug response as a function of tumor and features. However, most algorithm development efforts relied on cross-validation within single study assess model accuracy. While an essential first step, biological data set typically provides overly optimistic estimate the prediction performance independent test sets. provide more rigorous assessment generalizability between different studies, we...
Simultaneous localization and mapping (SLAM) plays a fundamental role in downstream tasks including navigation planning. However, monocular visual SLAM faces challenges robust pose estimation map construction. This study proposes system based on sparse voxelized recurrent network, SVR-Net. It extracts voxel features from pair of frames for correlation recursively matches them to estimate dense map. The structure is designed reduce memory occupation features. Meanwhile, gated units are...
Neural Network (NN) models provide potential to speed up the drug discovery process and reduce its failure rates. The success of NN requires uncertainty quantification (UQ) as explores chemical space beyond training data distribution. Standard do not information. Some methods require changing architecture or procedure, limiting selection models. Moreover, predictive can come from different sources. It is important have ability separately model types uncertainty, take assorted actions...
Abstract As renewable resources, such as wind, start providing an increasingly larger percentage of our energy needs, we need to improve understanding these so can manage them better. The intermittent nature the power generation makes it challenging for control room operators schedule wind while balancing load on grid. Forecasts be generated by a farm in hours ahead tend inaccurate, even under normal conditions. problem is exacerbated during ramp events, where changes large amount small...
The move toward exascale computing for scientific simulations is placing new demands on compression techniques. It expected that the I/O system will not be able to support volume of data written out. To enable quantitative analysis and discovery, we are interested in techniques compress high-dimensional simulation can provide perfect or near-perfect reconstruction. In this paper, explore use compressed sensing (CS) reduce size before they Using large-scale data, investigate how sufficient...
Wind energy is scheduled on the power grid using 0-6 hour ahead forecasts generated from computer simulations or historical data. When are inaccurate, control room operators use their expertise, as well actual generation previous days, to estimate amount of schedule. However, this a challenge, and it would be useful for have additional information they can exploit make better informed decisions. In paper, we techniques time series analysis determine if there motifs, frequently occurring...
Wind energy is scheduled on the power grid using 0–6 h ahead forecasts generated from computer simulations or historical data. When are inaccurate, control room operators use their expertise, as well actual generation previous days, to estimate amount of schedule. However, this a challenge, and it would be useful for have additional information they can exploit make better informed decisions. In paper, we techniques time series analysis determine if there motifs, frequently occurring diurnal...
Ramp events, which are significant changes in wind generation over a short interval, make it difficult to schedule energy on the power grid. Predicting occurrences of these events can help control room operators ensure that load and grid balance at all times. In this paper, we focus predicting up‐ramp large increases time interval. We propose novel detection algorithm uses historical data detect incoming pre‐ramp defined as part series occurs before ramp events. Using from Bonneville Power...
To enable personalized cancer treatment, machine learning models have been developed to predict drug response as a function of tumor and features. However, most algorithm development efforts relied on cross validation within single study assess model accuracy. While an essential first step, biological data set typically provides overly optimistic estimate the prediction performance independent test sets. provide more rigorous assessment generalizability between different studies, we use...
Compressing unstructured mesh data from computer simulations poses several challenges that are not encountered in the compression of images or videos. Since spatial locations points on a regular grid, as an image, it is difficult to identify near neighbors point whose values can be exploited for compression. In this paper, we investigate how three very different methods - spline fits, compressed sensing, and kernel regression compare terms reconstruction accuracy reduction size when applied...
The move toward exascale computing for scientific simulations is placing new demands on compression techniques. It expected that the I/O system will not be able to support volume of data written out. To enable quantitative analysis and discovery, we need techniques can compress high-dimensional simulation with near-perfect reconstruction. In this work, investigate Compressed Sensing (CS) reduce size from a fusion tokamak in 3 dimensions (Figure (a)). computational domain toroid, composed 32...
Gene expression profiles have been widely used to characterize patterns of cellular responses diseases. As data becomes available, scalable learning toolkits become essential processing large datasets using deep models model complex biological processes. We present an autoencoder capture nonlinear relationships recovered from gene profiles. The is a dimension reduction technique artificial neural network, which learns hidden representations unlabeled data. train the on collection tumor...
In this paper, we formulate the problem of predicting wind generation as one streaming data analysis. We want to understand if it is possible use weather in a time window just before current gain insight into how might behave interval after time. Specifically, singular value decomposition data, and that number values largest can be used predict magnitude change near future. The analysis uses an incremental algorithm based on sliding for reduced computational costs.
Neural Network (NN) models provide potential to speed up the drug discovery process and reduce its failure rates. The success of NN require uncertainty quantification (UQ) as explores chemical space beyond training data distribution. Standard do not information. Methods that combine Bayesian with address this issue, but are difficult implement more expensive train. Some methods changing architecture or procedure, limiting selection models. Moreover, predictive can come from different...