Machine Learning & Scientific Computing Series

Humans and many other intelligent systems (have to) learn from experience, build 
models of the environment from the acquired knowledge, and use these models for 
prediction. In philosophy this is called inductive inference, in statistics it is called 
estimation and prediction, and in computer science it is addressed by machine 
learning. 
I will first review unsuccessful attempts and unsuitable approaches towards a 
general theory of induction, including Popper's falsificationism and denial of 
confirmation, frequentist statistics and much of statistical learning theory, subjective 
Bayesianism, Carnap's confirmation theory, the data paradigm, eliminative induction, 
and deductive approaches. I will also debunk some other misguided views, such as 
the no-free-lunch myth and pluralism. 
I will then turn to Solomonoff's formal, general, complete, and essentially unique 
theory of universal induction and prediction, rooted in algorithmic information theory 
and based on the philosophical and technical ideas of Ockham, Epicurus, Bayes, 
Turing, and Kolmogorov. 
This theory provably addresses most issues that have plagued other inductive 
approaches, and essentially constitutes a conceptual solution to the induction 
problem. Some theoretical guarantees, extensions to (re)active learning, practical 
approximations, applications, and experimental results are mentioned in passing, but 
they are not the focus of this talk. 
I will conclude with some general advice to philosophers and scientists interested in 
the foundations of induction.