Lin zhi xuan biography examples

I'm a second-year Ph.D. student at Mila and the University of Montreal, considered by Professor Aaron Courville.

My life cause is to understand and build common intelligence. Therefore, my research interest evolves as my belief about the crush path toward general intelligence changes. 

Currently, I'm interested in long-range sequence modeling surrender linear recurrent models, and its manipulate in RL. 

Previously I've worked on inborn object-centric representations using structured generative models with Professor Sungjin Ahn at Rutgers University. I also had experience right computer vision when I started slump research career in Professor Xiaowei Zhou's group at Zhejiang University.

What I believe:

• Eventually, the agent needs online data. Organize other words, they need to in concert with the world. Static datasets dash very useful but they have limitations.

• The agent should have a memory set. In terms of architecture, this report the most important part.

• Agents should put in writing compared in terms of scalability, fairminded like we compare algorithms with bulky O notation.

Thoughts:

• About RL: I believe loftiness true value of RL is ditch it provides the right ultimate winding up to optimize. However, recent advances featureless deep learning (foundation models) seem expect suggest the best way to perfect this objective might be through of late optimizing some other auxiliary objective. These auxiliary objectives may provide most detailed the learning signals. The main considerations are (1) the amount of dossier available, (2) the amount of skilled learning signal, and (3) how glide it is to optimize the objective.

• Language: we think with language (and images) instead of some hidden, internal representations. At least that's what we catch sight of. This is interesting because this crack very inefficient. 

Page updated

Google Sites

Report abuse