Reading List of articles and notes from podcasts - doraithodla/notes GitHub Wiki
Podcast on Object Store
- Open table format
- Snowlake (#research)
- open json?
- When GPU becomes a commodity
- On good vs great products
The real value is in Applied AI
Reinventing transformers for the transfomer era
(Paused because I want to understand transformers well first)
Leveraging HCL for NLP
This was mostly on CogSci
-
Longer sentences have different properties
-
Intermediate representations (of sentences) of language
-
Human brain data from fMRI
-
Cognitive signals (brain signals)
-
Oxygen levels in the brain and its impact on brain signals (blood carries oxygen to the brain)
-
Using existing fMRI studies
-
Since different brain signals are of different intensity and have different noise levels , statistical pre-processing (I guess normalization) is a must before building models
-
Semantic similarity deduction by having the same language statements generating brain signals in different people)
-
Part of the research is focused on aligning concreate/abstract signals in brain for the same input data
-
Eye tracking research can be useful. It is over 50 years old
-
People fixate on more difficult words and skip over frequent as well as easy words (This has impact on learning through reading)
-
Fixation vs Attention
-
Interpretability data research can be used (though orthogonal) to aid research
-
Finding what humans do (in building brain models) and machines do (through ML) is another area of research
-
Emotion connection. Analyzing which word in a sentence corresponds to which emotion may yield interesting results
-
This can be done by masking words in a sentence
-
Gradient Salency
-
Reasonging over sentence boundaries