Fine tuning foundation models — The Dan MacKinlay stable of variably-well-consider’d enterprises
Fine tuning foundation models| The Dan MacKinlay stable of variably-well-consider’d enterprises
Event: ACTStats 2025 Annual Meeting Keynote Talk, Nashville TN USA Slides Details| Statistical Thinking
Fine tuning foundation models| The Dan MacKinlay stable of variably-well-consider’d enterprises
I don’t know much about this variant of Bayes, but the central idea is that we consider Bayes updating as a coherent betting rule and back everything else out from that. This gets us something like classic Bayes but with an even more austere approach to what probability is. I am interested in this because, following an insight of Susan Wei’s, I note that it might be an interesting way of understanding when foundation models do optimal inference, since most neural networks are best underst...| The Dan MacKinlay stable of variably-well-consider’d enterprises
Albergo, Boffi, and Vanden-Eijnden. 2023. “Stochastic Interpolants: A Unifying Framework for Flows and Diffusions.”| The Dan MacKinlay stable of variably-well-consider’d enterprises
Disentangled representation learning| The Dan MacKinlay stable of variably-well-consider’d enterprises
Neural denoising diffusion models of language| The Dan MacKinlay stable of variably-well-consider’d enterprises
1 Key Research Directions| The Dan MacKinlay stable of variably-well-consider’d enterprises
Figure 1 Jonathan Huggins summarizes: Complexity of Inference in Bayesian Networks. To cover: Sampling from a posterior measure versus calculating, it, approximation versus exact computation. Graphical models. What does calculation even mean on arbitrary measure spaces? 1 References Bodlaender, Donselaar, and Kwisthout. 2022. “Parameterized Complexity Results for Bayesian Inference.” Cooper. 1990. “The Computational Complexity of Probabilistic Inference Using Bayesian Belief Networks....| The Dan MacKinlay stable of variably-well-consider’d enterprises
Figure 1 I just ran into this area while trying to invent something similar myself, only to find I’m years too late. It’s an interesting analysis suited to relaxed or approximated causal modelling of causal interventions. It seems to formalise coarse-graining for causal models. We suspect that the notorious causal inference in LLMs might be built out of such things or understood in terms of them. 1 Causality in hierarchical systems A. Geiger, Ibeling, et al. (2024) seems to summarise SOT...| The Dan MacKinlay stable of variably-well-consider’d enterprises
Figure 1 Placeholder, for notes on what kind of world models reside in neural nets. 1 Incoming NeurIPS 2023 Tutorial: Language Models meet World Models 2 References Basu, Grayson, Morrison, et al. 2024. “Understanding Information Storage and Transfer in Multi-Modal Large Language Models.” Chirimuuta. 2025. “The Prehistory of the Idea That Thinking Is Modelling.”Human Arenas. Ge, Huang, Zhou, et al. 2024. “WorldGPT: Empowering LLM as Multimodal World Model.” In Proceedings of the ...| The Dan MacKinlay stable of variably-well-consider’d enterprises
Figure 1 Certifying NNs to be what they say they are. Various interesting challenges in this domain. I am not sure if this is well-specified category in itself. Possibly at some point I will separate the cryptographic verification from other certification ideas. Or maybe some other taxonomy? TBD 1 Ownership of models Keyword: Proof-of-learning, … (Garg et al. 2023; Goldwasser et al. 2022; Jia et al. 2021) TBD 2 Proof of training E.g. Abbaszadeh et al. (2024): A zero-knowledge proof of trai...| The Dan MacKinlay stable of variably-well-consider’d enterprises
Janice Pogue Lecture in Biostatistics, Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, Ontario, Canada 2024-12-06 Slides| Statistical Thinking
Background Consider these four conditions: There is no reliable prior information about an effect and an uninformative prior is used in the Bayesian analysis There is only one look at the data The look was pre-planned and not data-dependent A one-sided assessment is of interest, so that one-tailed p-values and Bayesian posterior probabilities are used, where is the effect parameter of interest (e.g., difference in means, log effect ratio) and means “conditional on” or “given”. One-Sid...| Statistical Thinking
Event: Consilium Scientific Slides Video| Statistical Thinking
I would like to write a small review of the book Bernoulli’s Fallacy by Aubrey Clayton. First the conclusion: this is a well researched and important book. My rating is a strong buy, and Bern…| Win Vector LLC
Just the other day1 I was chatting with a friend2 about MCMC and he asked me a fundamental, but seldom asked, question: What happens my acceptance probability is a bit off?. This question comes up a bunch. In this context, they were switching from double to single precision3 and were a little worried that some of their operations would be a bit more inexact than they were used to. Would this tank MCMC? Would everything still be fine? What is Markov chain Monte Carlo Markov chain Monte Carlo (...| Un garçon pas comme les autres (Bayes)
Sometimes it’s the parable of the barren fig tree. Sometimes you’re just pissed at a shrub. Paradoxes and counterexamples live in statistics as our morality plays and our ghost stories. They serve as the creepy gas station attendants that populate the roads leading to the curséd woods; existing not to force change on the adventurer, but to signpost potential danger.1 As a rule, we should also look in askance at attempts to resolve these paradoxes and counterexamples. That is not what the...| Un garçon pas comme les autres (Bayes)
Background As explained here, the power for a group comparison can be greatly increased over that provided by a binary endpoint, with greater increase when an ordinal endpoint has several well-populated categories or has a great many categories, in which it becomes a standard continuous variable. When a randomized clinical trial (RCT) is undertaken and deaths can occur, there are disadvantages to excluding the death and analyzing responses only on survivors using death as a competing risk, wh...| Statistical Thinking
UCLA Cardiology Grand Rounds 2020-10-23 | Video (better video below) Vanderbilt University Department of Biostatistics 2020-11-18 Vanderbilt Translational Research Forum 2021-11-04 | Video Consilium Scientific 2024-03-14 | Video and here Slides| Statistical Thinking
Background Consider the problem of comparing two treatments by doing squential analyses by avoiding putting too much faith into a fixed sample size design. As shown here the lowest expected sample size will result from looking at the developing data as often as possible in a Bayesian design. The Bayesian approach computes probabilities about unknowns, e.g., the treatment effect, and one can update the current evidence base as often as desired, knowing that the current information has made pre...| Statistical Thinking
Suppose that a patient is to be screened for a certain disease or medical condition. There are two important questions at the outset. How accurate is the screen or test? For example, at the outset,…| A Blog on Probability and Statistics
It's been more than a decade since Eliezer Yudkowsky started writing The Sequences. Lots of stuff has happened since then| www.thelastrationalist.com
