There's a lot of ML research on what functions need fitting. In particular, the field of inductive logic programming – learning logical theories – should be relevant. Also, to Konrad's point, I think many people working on causal learning see connections with "mainstream" ML.
MACHINE LEARNING
-
Causal Theories as Inductive Biases: Benefits and Questions
By
–
Of course, but the question is: how have causal theories to date benefited from being viewed as inductive biases?
-
Building Better ASR Models Community Event Discord
By
–
Brilliant idea. The event would be focused on building better ASR models but you can join our discord server and we have a thriving community of people experimenting and testing things out. They’ll be more than happy to help out.
-
Understanding Problem-Solving: Key to AI Mastery
By
–
when you figure out you can figure out anything, you figure out everything
-
Learning Algorithms and Inductive Biases in Action Variables
By
–
Essentially every learning algorithm / inductive bias is applicable to this. E.g., think of learning a Bayes net where some of the variables are your actions. (How well it works is another question, and here causal theories are presumably very useful, but not exhaustive.)
-
Evaluation Metrics for Language Modeling Explained
By
–
Evaluation Metrics for Language Modeling https://
thegradient.pub/understanding-
evaluation-metrics-for-language-models/
… -
Shannon’s Mathematical Theory of Communication Foundations
By
–
A Mathematical Theory of Communication By C. E. SHANNON
-
Physics Laws as Master Algorithm: Intelligence Emergence
By
–
The laws of physics are the master algorithm: run them long enough and intelligence emerges.
-
Inductive Bias, Complete Data, and Causal Theories Recap
By
–
To recap:
– We always need prior assumptions/inductive biases, unless we have complete data (no free lunch theorem).
– If we have complete data, we don't need anything else for any rung.
– Causal theories are a type of inductive bias.
I guess that's a No answer to your question. -
Incomplete Models: Action’s Role in Learning and Decision-Making
By
–
Those models are incomplete, because they don't include your actions. Convergence aside, and tautologically, if you know everything there's nothing left to learn. You can then make optimal decisions without causal assumptions.