A scratch-built neural network training and application suite. I implemented a customizable deep-neural network to train against randomized vector data to begin to experiment with video compression, wherein input seed vectors are smaller than the output vectors. I built a Genetic Algorithm as the primary training algorithm but also inluded experimental implementations of Particle Swarm Optimization and Ensemble Learning. The Genetic Algorithm is multi-threaded and I also attempted to harness GPU matrix evaluation using Aparapi but never completed it. The Genetic Algorithm manipulates the weights as an analog to DNA. At each round the DNA is evaluated by applying its properties to a network and running that network against the training data. Experiments can be run by changing which enum constants are used in Driver and by changing parameters in the training algorithms. Feel free to experiment and explore!
DeepMVS: Learning Multi-View Stereopsis
Always sparse. Never dense. But never say never. A Sparse Training repository for the Adaptive Sparse Connectivity concept and its algorithmic instantiation, i.e. Sparse Evolutionary Training, to boost Deep Learning scalability on various aspects (e.g. memory and computational time efficiency, representation and generalization power).
Neural networks lectures for computer science students.
Explore and create ML datasets. Sample the dataset and create training, validation, and testing datasets for local development of TensorFlow models. Create a benchmark to evaluate the performance of ML. TensorFlow is used for numerical computations, using directed graphs. Getting started with TensorFlow. Explore the TensorFlow python API, build a graph, run a graph, feed values into a graph. Find areas of a triangle using TensorFlow. Learning from tf.estimator. Read from python’s pandas dataframe into tf.constant, create feature columns for estimator, perform linear regression with tf.Estimator framework. Execute Deep Neural Network regression. Use benchmark dataset. Refactoring to add batching and feature creation. Refactor the input. Refactor the way the features are created. Create and train the model, Evaluate the model. Distributed training and monitoring. Create features out of input data. Train and evaluate. Monitor with Tensorboard. To run TensorFlow at scale, use Cloud ML Engine. Package up the code. Find absolute paths to data. Run the python module from the command line. Run locally using GCloud. Submit training job using GCloud. Deploy model. Make predictions. Train on a 1-million row dataset. Feature Engineering. Working with feature columns. Adding feature crosses in TensorFlow. Reading data from BigQuery. Creating datasets using Dataflow. Using a wide-and-deep model.
Longsword Stance Model Training Working demo: https://www.youtube.com/watch?v=v7hvOyPQ0EM Longsword Stance Model Training: Deep Learning model & training python scripts. The model is genenerated with Keras, as a multivariate Bidirectional Long-Short Term Memory (LSTM) network, for classification of longsword movement gestures. Softmax is being used as the activation function and sparse-categorical cross-entropy on the final dense layer
java deep learning algorithms and deep neural networks with gpu acceleration
This is the source code of large-scale bi-directional generative adversarial network (BigBiGAN) for deep learning semantic feature extraction.