Ideology

With every try, you have explored the space a little bit more, grown the subtree a little bit deeper, and pushed the "table" a little bit wider. Success never comes from one good state but rather the path you have explored and the large subtree you have built. The tree has been explored and nothing is lost.

Searching for Needles in Haystacks

To some extent, machine learning is all about feature engineering or feature learning \(\vec \phi(\vec x)\), trying to find a good/correct mathamatical space (may be very complex, may involve going to the weight world to find correct projection or may involve projecting to infinite dimension to find representation), a feature space \(\Phi\), to allow \(X\) project onto, which might be a process that is "searching for needles in the haystacks".

"Intelligence is just happening to do the right thing for the right task and engineering/creating something that works for an interestingly scaled level problem is the same as parsing through the fog in this vast space of interactions that give so many seemingly correct functions and finding the truth."

Inspired By professor Gao from UCSD: rigrous engineering is the act of walking in the infinite function space in the intersection between design space and the domain to try to find a good enough solution that can solve interesting problems when scaling up for the metrics you care about and under the constrain given. In some sense, search = optimization.

Inspired by professor Eldridge from UCSD: the math itself for learning to occur may not be as complicated, it is rather the underlaying structure of the data that exist in nature promotes the stemming of intelligence, we just need to find such structure, a correct representation of \(X\).

I am pashinate in the "science part" of doing research, to use imaginations to create learnings, to create different ways for an artificial agent to capture essence computationally, and to create something that seems to be "intelligent", that just happens to do the right things that works. Sometimes these approaches are very ideal, naive, but are also new and may have the capability to develop into something captures the truth.

Nature of Information Processing & Sequential Decision Making That Makes "Intelligence"

I believe that, no matter for natural or artificial intelligence, there exist, in nature, certain "correct" ways of information processing and decision making mechanism that makes "intellience". It just happens that the random search conducted by evolution found such mechanism and presented in the form of our brain. As long as natural intelligence remains as the best form of intellignece, it keep inform us about hwo to get closer to true mechnism of intelligence.

Inspired by professor Pereira from the Salk Institute: the nature has creative solutions to match objective functions caused by evolution and there exists a strong coupling between such natural behavior and the underlying neural algorithm. The neural underpinning happens to be the information processing theory that works because it helps survival to pass on the genes. Ethology is an integration of all computations done from sensory to supervisory signals. A great deal of neuronal mechanism is about our output, our actions, and our interaction with the environment: it is about what we do.

However, how we theorize these mechanisms may be limited to our imagination and natural mechanisms are not necessarily bounded within these constraints, they often rely on more complex features that only imperfectly map onto human-interpretable categories. I think that using the computational model with the right layer of abstractions would go beyond the limits of human-interpretable labels and potentially establish find natures of how information are processed and decisions are made, hence, benifiting both community of machine learning and neurosciecne.

VNL Schematics (borrowed from VNL Research Strategy)

Because of such fundamental goals of such neural algorithms, it helps to inject the basis layer of alignment with biology into artificial agents to form the base of embodiment in these networks and power searches into the correct representations to inform us and allow us to build abstract models of the brain, to get closer on the conceptual level truth of information processing & sequential decision making, and to find hypothesis to “intelligence”.