Today's Progress: 1. Studied Keras from a book: "Deep Learning with Python" by Francois Cholet. 2. Read an article titled 7 steps to learning Keras: https://www.kdnuggets.com/2017/10/seven-steps-deep-learning-keras.html 3. Watched a video (from pydata) on Keras: https://www.youtube.com/watch?v=FrkYu2zVUyM
Thoughts: I am beginning to grasp the function class / Model class. See this documentation for more info: https://keras.io/models/about-keras-models/
Today's Progress: 1. Studied Keras from a book: "Deep Learning with Python" by Francois Cholet. Today I finished reading the last chapter. 2. Watched the first 30 minutes of a more recent video (from pydata) on Keras: https://www.youtube.com/watch?v=BBIA6Wcu2j4 3. Bookmarked a set of keras implementations of Generative Adversarial Networks
Thoughts: Today I learned about the implementation of 'temperature' in a RNN/LSTM. I was pleased to learn that keras functionality is low enough level to accomplish a dynamic temperature. My hope is to reproduce (in keras) the results of this blog: https://www.mtgsalvation.com/forums/magic-fundamentals/custom-card-creation/612057-generating-magic-cards-using-deep-recurrent-neural?page=32
Today's Progress: 1. Watched the last hour of the same video from yesterday. 2. Watched Siraj Raval's video on how to teach AI: https://www.youtube.com/watch?v=tczjZOLVjJM&t=654s
Thoughts: One term stuck out to me that I failed to understand: Batch Normalization. I'm reading this article to amend that: https://towardsdatascience.com/batch-normalization-in-neural-networks-1ac91516821c
Today's Progress: 1. Watched the first sixteen episodes of 'AI Adventures' https://www.youtube.com/watch?v=0uXMgLIlXoE&list=PLIivdWyY5sqJxnwJhe3etaK7utrBiPBQ2&index=17 2. Watched two videos by sentdext on Tensorboard and model optimization. 3. Printed out and read over the original GAN paper by Ian Goodfellow, Yoshua Bengio, Sherjil Ozair, and others. 4. Wrote out some basic vocabulary terms that every beginner should familiarize themselves with. In the next few weeks I will be helping to organize a meetup in San Francisco. I did this work as preperation for a handout I could share with students to make sure everyone is caught up as best as possible
Today's Progress: 1. reread the original GAN paper 2. Printed and read the paper 'Improved Techniques for training GANs' https://arxiv.org/pdf/1606.03498.pdf 3. tried out the code for the first 5 chapters of "Deep Learning with Python" by Francois Cholet. https://github.com/fchollet/deep-learning-with-python-notebooks
Thoughts: When I tried to run some of Francois Cholet's code on CNNs today, I found it estimating each epoch at 5+ hours... so I shut down training and checked the GPU. Of course I'd forgotten to activate the environment in which I have the GPU version of tensorflow and cuDNN properly installed (whoops). After activating, each epoch took about ten seconds. I found this to be a far more satisfactory wait time.
Today's Progress: 1. printed and read a paper titled: 'Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks' 2. reread the article titled 'fantastic GANs and where to find them', as well as the second section of this article: http:https://guimperarnau.com/blog/2017/03/Fantastic-GANs-and-where-to-find-them 3. Learned about IBM Watson helping to identify bad actors in trading: https://www.foxbusiness.com/features/ibm-tests-watson-technology-to-keep-eye-on-traders
Thoughts: Two lofty objectives: 1. Use artificial intelligence to augment human cognition 2. Improve the cognition of institutions, with the goal of realizing enlightened society
Today's Progress: 1. Today I broke my computer. That is, I broke the connection between Tensorflow and CUDA. I don't know how I did it, except that I expect it had something to do with updating tensorflow. I tried uninstalling tensorflow and I got an error. I deleted the folder and reinstalled and found myself with more errors. I created about 4 or 5 new environments with different versions of tensorflow and a bunch of new problems. The end state has me scratching my head. When I open idle from anaconda, tf.test.is_built_with_cuda() returns true... but when I open a jupyter notebook (from the same environment) this returns false.
Thoughts: Considering giving up entirely. I have no one to call about this issue, and I'm not willing to reinstall everything from scratch (yet). Extremely depressed.
Today's Progress: 1. Studying Math today.
Thoughts: Next thing to try (GPU fix): reinstall anaconda from scratch