Day #193 - Deep Learning State of the Art (2019) - MIT

Summary of AI progress and important milestones in 2018. Good Summary Talk.

Key Summary

• Breakthrough in 2017 /18

• RNN Sequence Models
• Encode -> Sequence Map to Fixed Size Vector
• Forms Representation, Feeds Representation in Decoder Sentence

• Attention - Mechanism to look back at Input Sequence as part of decoding process
• Decoder interprets the hidden state

• Word Embeddings  (Word Sequence following order)

• Elmo Embedding, Bert applications

• Tesla Autopilot Overview

• AutoML - Architecture based on data

• AutoAugmentation - Technique to Augment Data (Learn a Lot from Little)

• Training with Randomized Data

• Segmentation / Bounding box detection in image

• RPN Networks / Single Shot methods
• SSD / Region based methods

• TL - Transfer weights learned on a task and finetune to next level of dataset
• DL Frameworks

These two slides are motivating

Happy Mastering DL!!!