Get Started
image

image Weekly Tech News – Week #7

image DeepETA

Uber has moved from XGBoost to Deep Neural Networks. Check out how their new model "DeepETA" drastically improves the expected arrival times (ETA) of uber drives in their latest blog published here.

image MuZero

From the creators of AlphaGo, DeepMing has produced a new approach towards online streaming compression. The approach claims to reduce around 4.7% bitrate which is a sizable contribution given that majority of the internet traffic is video. Checkout the detailed description here.

image OPEN AI

If you are interested in CLIP models, here is a great news: OpenAI stealth released the model weights for the largest CLIP models, RN50x64 & ViT-L/14. Check out their tweet here.
#MachineLearning #generativeart #vqganclip #generative #AI

image PhD thesis: On Neural Differential Equations

For those pursuing their careers in the field of neural differential equations (NDEs), Patrick Kidger, a Postgraduate Research Student from University of Oxford, has published his PhD thesis on archives, check out the link here. The thesis should be of particular interest of those seeking the overlap between deep learning with dynamical systems. The topics covered in this thesis includes neural ordinary differential equations, numerical methods for NDEs, symbolic regression for dynamical systems, and deep implicit models such as deep equilibrium models and differentiable optimisation.

image PGMax

PGMax is the python library for probabilistic graphical models. It implements general factor graphs for discrete probabilistic graphical models (PGMs), and hardware-accelerated differentiable loopy belief propagation (LBP) in JAX. Check out their tutorial using Colab on getting started with PGMax. Library can be directly installed via GitHub from here. Here is their official website.

image DIAMBRA

An AI Tournaments Platform , where every coder will be able to train his agents and compete. The platform will provide one-to-one fights against other agents and matches versus human players. It is based on reinforcement learning and built with OpenAI Gym Python interface, easy to use, transforms popular video games into Reinforcement Learning environments. Check out their latest news here and GitHub code here.

image FHEz

Python-FHEz is a fully-homomorphically-encrypted deep-learning library. This library supports a lot of primitive in the area of deep learning where data is high confidential and private. Homomorphic encryption allows running certain calculations in an encrypted fashion or transfer information in an enycrpyted way such that the involved parties would not know the contents of the data. Check out their GitHub code here.

image TorchMetrics

TorchMetrics introduce a general-purpose metrics package that covers a wide variety of tasks and domains used in the machine learning community. It provides standard classification and regression metrics as well as domain-specific metrics for audio, computer vision, natural language processing, and information retrieval. Link to their original paper is here and their GitHub repo can be found here.

Gym

Gym is a standard API for reinforcement learning, and a diverse collection of reference environments. After being 5 months in maintenance, Gym has now finally announced it proper documentation that is to be found here. It is still in its beta version though and any issues can be directly reported in their GitHub repo here.