sam-lev
CS PhD student @ the Univ. of Utah / Scientific Computing & Imaging Institute with a focus on machine learning, topological computing & their novel combination.
University of Utah
Pinned Repositories
2016-dataviscourse-homework-Sam-Leventhal
Artificial-Intelligence-Berkeley-CS188
Learned about search problems (A*, CSP, minimax), reinforcement learning, bayes nets, hidden markov models, and machine learning
Classifications-Of-Songs-via-Homology-of-Chroma-Features
Homological features are identified through persistent homology of song chroma features whose persistence diagrams serve as a signature for song clustering. Clustering algorithms emplyeed are k-Means and k-Median.
Convolutional-Nueral-Net-For-3D-Protein-Renderings
Classifies images of proteins by family taken from 3D models via a convolutional nueral net to encode images into a feature vector used as input by a long short term memory nueral net.
Convolutional_Neural_Net_Lymphoma
Deep, dense convolutional neural net to discern between different Lymphoma types (Burkitt and Diffuse Large B-Cell) from images of patient tissue samples.
HPC_ML_Hierarchical_IO
neural-net-path-tracer
conditional generative advisarial network trained on rendered images and conditional image buffers to automate path traced images and global illumination using vtk-m and pytorch
Spectral-Graph-Set-Similarity-Analysis-For-Quantum-Structuring-In-Keplerian-Systems
topological-distortion-of-dimension-reduction
UnSupUnetMSCSegmentation
unsupervised u-neural net using morse smale complex topological segmentation at various persistences to learn segmentation of optic vessel images. Contribution is the lack of needed hand segmentation used in the original Unet paper by Ronneberger et. al. via topological training set.
sam-lev's Repositories
sam-lev/Classifications-Of-Songs-via-Homology-of-Chroma-Features
Homological features are identified through persistent homology of song chroma features whose persistence diagrams serve as a signature for song clustering. Clustering algorithms emplyeed are k-Means and k-Median.
sam-lev/neural-net-path-tracer
conditional generative advisarial network trained on rendered images and conditional image buffers to automate path traced images and global illumination using vtk-m and pytorch
sam-lev/Convolutional_Neural_Net_Lymphoma
Deep, dense convolutional neural net to discern between different Lymphoma types (Burkitt and Diffuse Large B-Cell) from images of patient tissue samples.
sam-lev/HPC_ML_Hierarchical_IO
sam-lev/UnSupUnetMSCSegmentation
unsupervised u-neural net using morse smale complex topological segmentation at various persistences to learn segmentation of optic vessel images. Contribution is the lack of needed hand segmentation used in the original Unet paper by Ronneberger et. al. via topological training set.
sam-lev/2016-dataviscourse-homework-Sam-Leventhal
sam-lev/Artificial-Intelligence-Berkeley-CS188
Learned about search problems (A*, CSP, minimax), reinforcement learning, bayes nets, hidden markov models, and machine learning
sam-lev/Convolutional-Nueral-Net-For-3D-Protein-Renderings
Classifies images of proteins by family taken from 3D models via a convolutional nueral net to encode images into a feature vector used as input by a long short term memory nueral net.
sam-lev/Spectral-Graph-Set-Similarity-Analysis-For-Quantum-Structuring-In-Keplerian-Systems
sam-lev/topological-distortion-of-dimension-reduction
sam-lev/examples
A set of examples around pytorch in Vision, Text, Reinforcement Learning, etc.
sam-lev/GraphSAGE
Representation learning on large graphs using stochastic graph convolutions.
sam-lev/Markov-Chain-Based-File-Replicator
An implemented a Nondeterministic Finite Automaton to replicate various literary works where our program output is based on stochastic production laws designed according to chain frequencies within any given text file. For the purpose of example our production schematic is used to produce replicas of such stories as To Kill a Mockingbird and multiple Shakespearean plays. The purpose of this write up is then to demonstrate how this automaton was implemented, what possible alterations can be made to achieve other simple computational models, and the possible advantages/disadvantages they may provide in trying to replicate human writing. More specifically it is shown that our Nondeterministic Finite Automaton, based on Markov chain production rules can be interpreted as a context free grammar and can be easily altered into a Context Sensitive Grammar for various applications or improvement in certain areas.
sam-lev/NeuralMT
FBK Neural Machine Translation Toolkit with instance-based adaptation
sam-lev/PAVE_ISAV
Journal Submission for PAVE project at ORNL with Mark Kim and Dave Pugmire
sam-lev/sam-lev.github.io
sam-lev/signedDistanceField_2D_Images
Computes the signed distance field over a 2D image using the Gaussian mean of the first channel as the isovalue.