A collection of video talks and recorded live demos about the 1 million neuron TrueNorth chip, a custom computer processor that implements neural networks extremly efficiently and scalably. Talks range from big picture thoughts on the space of computer science to vision applications running on the chip, production programming methodologies for neural networks, robust asynchronous circuit design, and silicon retinas that will revolutionize sight beyond CCDs.
Bayesian approach avoiding observed overfitting from cross validation with low data by collecting global information of all tests and then projecting uncertainty to the individual models with emperically benifits presented.
Data analysis drives machine learning model to personalize search results that benifit both the guests and hosts in a two-sided marketplace. A standard Collaborative Filtering approach did not work due to noisy data, instead they smoothed the data and used logistic regression with scikit-learn to achieve ~4% improvement in booking over 30 days.
Think deep learning doesn’t have enough foundational theory? Love the warm blanket of statistical machine learning? It turns out denoising autoencoders can be interpreted as an instance of pseudo-likelihood learning.
What skills are necessary to develop to become a data scientist? Or the dual problem, what skills should we interview for hiring a data scientist? This is a short post describing some of the high level features: coding skills, ability to understand data, curiosity, imagination, explaining results in simple english, and an area of technical expertise.