Categories

Classification

Training with a large number of classes

In machine learning we often face the issue of a very large number of classes in a classification problem. This causes a bottleneck in the computation. There’s though a simple and eff...

In Classification, Mar 15, 2017
Training with a large number of classes

Optimization

Training Models that have Zero Likelihood

How can we train generative models when we cannot calculate a gradient likelihood?

In Optimization, Dec 21, 2019
Training Models that have Zero Likelihood

Some modest insights into the error surface of Neural Nets

Did you know that feedforward Neural Nets (with piecewise linear transfer functions) have no smooth local maxima?

In Optimization, Jul 30, 2017
Some modest insights into the error surface of Neural Nets

Evolutionary Optimization as a Variational Method

How popular evolutionary style optimisation methods can be viewed as optimisation of a rigorous bound.

In Optimization, Apr 03, 2017
Evolutionary Optimization as a Variational Method

Reinforcement Learning

Learning From Scratch by Thinking Fast and Slow with Deep Learning and Tree Search

According to dual process theory human reasoning consists of two different kinds of thinking.System 1 is a fast, unconscious and automatic mode of thought, also known as intuition. Sy...

In Reinforcement Learning, Nov 07, 2017
Learning From Scratch by Thinking Fast and Slow with Deep Learning and Tree Search

Natural Language Processing

Generative Neural Machine Translation

Machine Learning progress is impressive, but we argue that progress is still largely superficial.

In Natural Language Processing, Sep 12, 2018
Generative Neural Machine Translation

University College London

UCL AI Centre

The AI Centre carries out foundational research in AI. As we transition to a more automated society, the core aim of the Centre is to create new AI technologies and advise on the use ...

In University College London, Jan 18, 2019
UCL AI Centre

General

AI Coworkers

We argue for a more interactive approach in which AI systems function more like coworkers. For them to be effective in this role, they need to have reasonable estimates of confidence ...

In General, Jun 01, 2020
AI Coworkers

Featured

  1. AI Coworkers
    In General,
  2. Training Models that have Zero Likelihood
    In Optimization,
  3. Learning From Scratch by Thinking Fast and Slow with Deep Learning and Tree Search
    In Reinforcement Learning,
  4. Some modest insights into the error surface of Neural Nets
    In Optimization,
  5. Evolutionary Optimization as a Variational Method
    In Optimization,
  6. Training with a large number of classes
    In Classification,