TL;DR
This paper introduces a double-Bayesian framework for decision-making in machine learning, revealing intrinsic uncertainty and explainability, and linking solutions to the golden ratio through fixed points of Bayesian processes.
Contribution
It proposes a novel double-Bayesian approach that models decision-making as two intertwined Bayesian processes, offering new insights into uncertainty and solution characterization.
Findings
Decision-making involves dual Bayesian processes.
Solutions relate to fixed points and the golden ratio.
Framework aligns with neural network training parameters.
Abstract
Contemporary machine learning methods will try to approach the Bayes error, as it is the lowest possible error any model can achieve. This paper postulates that any decision is composed of not one but two Bayesian decisions and that decision-making is, therefore, a double-Bayesian process. The paper shows how this duality implies intrinsic uncertainty in decisions and how it incorporates explainability. The proposed approach understands that Bayesian learning is tantamount to finding a base for a logarithmic function measuring uncertainty, with solutions being fixed points. Furthermore, following this approach, the golden ratio describes possible solutions satisfying Bayes' theorem. The double-Bayesian framework suggests using a learning rate and momentum weight with values similar to those used in the literature to train neural networks with stochastic gradient descent.
Peer Reviews
Decision·Submitted to NeurIPS 2024
The authors are exploring an idea which is novel, and the whole thinking about Bayes classifiers as comprising two sub-problems seems novel and worth pursuing.
I did not really understand the discussion with the vase, the sender and receiver. Perhaps the authors should somehow connect the Bayesian ideas to the description of the problem earlier? I think the paper would really benefit from rewriting Section 4 with the vase as a running example, because it is hard to connect the various decisions with the probabilities. Maybe it's worth to add more illustrations / diagrams for this? The authors are presenting novel ideas and it's hard to understand them
* The paper takes a fresh look at decision marking under uncertainty, which is at the center of machine learning. * The generality of the setting makes the discussion applicable to virtually all of ML.
While I am sensible to the topic of prior and model elicitation from coherence arguments, I believe the paper needs a thorough revision focussing on clarity. While I have some intuition now, it is still not crystal clear to me what the exact goal or claims of the paper are. See bullets below for constructive comments. ## Major 1. Section 4: what is the probability $P$? What is the underlying space and sigma algebra? What are they supposed to represent? 2. Section 4 introduces several very str
Given that I don't properly understand what exactly the authors want to achieve, I am unable to formulate the strengths of this paper.
The presentation is very messy. The paper jumps from topic to topic without me understanding their relations to each other.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMachine Learning and Algorithms
MethodsBalanced Selection