Stochastic Re-weighted Gradient Descent (RGD)

#11: AnyMAL from Meta, and more..

Stochastic Re-weighted Gradient Descent (RGD)

• RGD introduces a variant of the SGD algorithm that re-weights data points during each optimization step based on their difficulty via Distributionally Robust Optimization

• This method uses the loss as a proxy to calculate the difficulty of a point, and re-weight it by the exponential of its loss.

• Read more on the blog here and paper here.

Image from the Blog

Any-Modality Augmented Language Model (AnyMAL)

• A unified model that reasons over diverse input modality signals (i.e. text, image, video, audio, IMU motion sensor), and generates textual responses.

• It converts modailty-specific signals to the joint textual space through a pre-trained aligner module. The model demostrates SOTA performance on various multimodal tasks.

• Read the paper here.

Image from the Paper

And More..

• International Conference on Computer Vision - ICCV, is happening in Paris from Oct 2 to Oct 6, 2023.

• Today is the last date to apply for AI Demo Day happening in Bengaluru on Oct 7.