Browsing by Subject "Transfer Learning"

Now showing 1 - 4 of 4
  • Thumbnail Image
    Item
    Detecting Biomarkers among Subgroups with Structured Latent Features and Multitask Learning Methods
    (2017-05) Zhang, Huanan
    Because of disease progression and heterogeneity in samples and single cells, biomarker detection among subgroups is important as it provides better understanding on population genetics and cancer causative. In this thesis, we proposed several structured latent features based and multitask learning based methods for biomarker detection on DNA Copy-Number Variations (CNVs) data and single cell RNA sequencing (scRNA-seq) data. By incorporating prior known group information or taking domain heterogeneity into consideration, our models are able to achieve meaningful biomarker detection and accurate sample classification. 1. By cooperating population relationship from human phylogenetic tree, we introduced a latent feature model to detect population-differentiation CNV markers. The algorithm, named tree-guided sparse group selection (treeSGS), detects sample sub- groups organized by a population phylogenetic tree such that the evolutionary relations among the populations are incorporated for more accurate detection of population- differentiation CNVs. 2. We applied transfer learning technic for cross-cancer-type CNV studies. We proposed Transfer Learning with Fused LASSO (TLFL) algorithm, which detects latent CNV components from multiple CNV datasets of different tumor types and distinguishes the CNVs that are common across the datasets and those that are specific in each dataset. Both the common and type-specific CNVs are detected as latent components in matrix factorization coupled with fused LASSO on adjacent CNV probe features. 3. We further applied multitask learning idea on scRNA-seq data. We introduced variance-driven multitask clustering on single-cell RNA-seq data (scV DMC) that utilizes multiple cell populations from biological replicates or related samples with significant biological variances. scVDMC clusters single cells of similar cell types and markers but varies expression patterns across different domains such that the scRNA-seq data are adjusted for better integration. We applied both simulations and several publicly available CNV and scRNA-seq datasets, including one in house scRNA-seq dataset, to evaluate the performance of our models. The promising results show that we achieve better biomarker prediction among subgroups.
  • Thumbnail Image
    Item
    Improving Automatic Painting Classification Using Saliency Data
    (2022-10) Kachelmeier, Rosalie
    Since at least antiquity, humans have been categorizing art based on various attributes. With the invention of the internet, the amount of art available and people searching for art has grown significantly. One way to keep up with these increases is to use computers to automatically suggest categories for paintings. Building upon past research into this topic using transfer learning as well as research showing that artistic movement affected gaze data, we worked to combine transfer learning with gaze data in order to improve automatic painting classification. To do this, we first trained a model on a large object recognition dataset with synthesized saliency data. We then repurposed it to classify paintings by 19th century artistic movement and trained it further on a dataset of 150 paintings with saliency data collected from 21 people. Training on this was split into two stages. In the first, the final layer of the model was trained on the dataset with the rest of the model frozen. Next, the entire model was fine-tuned on the data using a much lower learning rate. Fifteen trials of this were done with different random seeds in order to decrease any effect that randomness might have. Overall it achieved an accuracy of 0.569 with standard deviation of 0.0228. Comparatively, a similar existing method had an accuracy of 0.523 with standard deviation of 0.0156. This ends up being a statistically significant difference (p = 0.0479), suggesting that when given enough training time a more complex model utilizing saliency data can outperform a simpler model that does not use saliency data when it comes to classifying paintings.
  • Thumbnail Image
    Item
    Robotic Embodiment of Human-Like Motor Skills via Sim-to-Real Reinforcement Learning
    (2021-12) Guzman, Luis
    State of the art methods continue to face difficulties automating many tasks, particularly those which require human-like dexterity. The proposed "Internet of Skills" enables robots to learn advanced skills from a small set of expert demonstrations, bridging the gap between human and robot abilities. In this work, I train Reinforcement Learning (RL) control policies for the tasks of hand following and block pushing. I build a sim-to-real pipeline and demonstrate these policies on a Kinova Gen3 robot. Lastly, I test a prototype system that allows an expert to control the Kinova robot using only their arm movements, captured using a Vicon motion tracking system. My results show that performance of state of the art RL methods could be improved through the use of demonstrations, and I build a shared representation of human and robot action that will enable robots to learn new skills from observing expert actions.
  • Thumbnail Image
    Item
    Segmentation and Dense Keypoints Estimation of Monkeys
    (2021-12) Yu, Haozheng
    Animal tracking and pose estimation are core topics in neuroscience. However, for monkeys, current deep learning based algorithms often fail to perform well on segmentation and dense keypoints estimation due to the lack of annotated training data. In this thesis, we address this challenge by developing transfer learning based deep learning algorithms without using fully-annotated monkey data. We develop a bootstrapping strategy to refine the pretrained segmentation model on monkey data annotated with 2D sparse landmarks. In addition, we implement a voxel-based visual hull reconstruction approach to recover the 3D monkey pose from the silhouettes. For dense keypoints estimation, we follow a similar bootstrapping strategy to refine a pretrained HRNet, which is then used to learn a dense keypoint detector by leveraging multiview consistency. Our methods outperform the baseline methods on in-the-cage and in-the-wild monkey data.