Every week, my team at Invector Labs publishes a newsletter to track the most recent developments in AI research and technology. You can find this week’s issue below. You can sign up for it below. Please do so, our guys worked really hard on this:
From the Editor: The Friction Between Interpretability and Accuracy in Deep Learning
One of the biggest challenges of building deep learning models is trying to understand how they arrive to conclusions. The emergence of deep learning increased the level of complexity of traditional machine learning models by a significant multiple. Today, it is common to encounter fairly simple neural networks with millions of nodes and hundreds of hidden layers. Navigating those complex structures to interpret the decisions of a model is nearly impossible. Deep learning theory often refers to this phenomenon as the accuracy-interpretability friction.
The idea of the accuracy-interpretability friction is very simple. Models that are simple to interpret tend to not perform well in sophisticated environments while more robust models are nearly impossible to interpret. As deep learning evolves, there have been an increasing need to develop tools that improve the interpretability and visualization of deep learning models. This week, IBM released a new toolkit with that sole purpose. Certainly, interpretability will be at the center of the next decade of deep learning innovation.
Now let’s take a look at the core developments in AI research and technology this week:
MIT unveiled a new machine learning models that automates the annotation of massive datasets for medical research.
Google published a paper outlining a technique known as temportal-cycle consistency learning, which can simplify the training of video analysis models.
DeepMind disclosed some of their recent work in deep learning models and datasets for ecological research.
Cool AI Tech Releases
IBM released AI Explainability 360, a new toolkit to improve the interpretability of machine learning models.
Microsoft and Carnegie Mellon University announced the MineRL competition which looks to leverage Minecraft-based Project Malmo to advance reinforcement learning solutions.
Google released EfficientNet-EdgeTPU, a family of image classification models based on AutoML and optimized for Google’s Edge TPU.
AI in the Real World
U.K.’s National Health Service (NHS) is building a new unit to tackle AI health care challenges.
The US intelligence revealed a project called Sentient that has been described as an artificial brain.
MIT published an interesting analysis about China’s brain drain when comes to AI talent.