107 - Multi-Modal Transformers, with Hao Tan and Mohit Bansal
NLP Highlights
00:00
The Intuition Behind Object Recognition Returning Tasks
The motivation is that we want, i motive to capture bost aliable information and the low lav information. The low lave information is captured by the future agression that will just directed one to request the to tesons. And the forty eighth dimensional tans feture of the resinat so it would capture the information such as the color and the texture of the objects. We want to capture the la information. So this, irosate kind of information. Do you actually see that the way the model learns is different for these two tasks? And so you distubed an intuition. Did you actuly see that in your results? Is i think i could observe that,
Transcript
Play full episode
