I am collaborating with Professor Buxton at the University of Illinois (Springfield) to develop a language model to determine document similarity between hundreds of documents in two groups. We are exploring NLP models (BERT, Longformer, SBert) to determine which model is best at technical (scientific) document similarity.
The project uses Python and Pytorch to load data from two different sets of files (A and B) to test which files from 'A' match most closely with the files from 'B'. I wrote custom classes to load and clean the data. The code runs on Google's Colab to finetune existing models with the specific technical files.
Helpful explanations about the nuts and bolts of NLP.
This book helped me see that AI (computerized intelligence) is not capable of coming close to human intelligence. Organic intelligence and digitized intelligence are two different things. Computer will not be able to achieve Artificial General Intelligence (AGI).
This book was enjoyable because it takes a contrarian approach to AI research. It is philosophical and touches upon multiple disciplines (psychology, neuro-science). The author makes a clear case that AI isn’t close to understanding how humans think. We have an intelligence that cannot be replicated through "deduction" nor "induction." Deduction is using logic to solve problems given information that already exists. It does not add new knowledge, it only finds hidden knowledge. Induction is predicting future events based on the statistics of past events. For example, if I always pull a white marble from a bag, then I could declare that all the marbles in the bag are (probably) white.
But true AI needs to be completely generalizable. This requires "Abduction" which is generating new ideas, thoughts, summaries, that go beyond the existing data. It makes a leap to what might be. It is not always accurate, but it can lead to break-through discoveries.
I propose that we will never be able to "create life" like God can. So my attempts to create an application (using human intelligence) will include symbolic AI (to build up understanding block by block) along side statistical (learning) approaches.
I used this book for my deep learning class at UIS.
I used this book for my deep learning class at UIS.
I used this book for my deep learning class at UIS.
I will continue NLP and AI research to help with the creation of a smart game. A game is smart if it can learn enough about its players to teach the players something about their life purpose.