Recent content on Vadim Liventsev
Consider the following search problem: The above elevation map of Vaalserberg (the hill that acts as a three-way border between Belgium, the Netherlands and Germany) contains a telecommunications tower. Can you find it? Hint: in a hilly terrain, telecommunication towers work better when they are higher up. If you are anything like me, you would start by looking at the highest points on the map. If…
Ladies and gentlemen of the internet, consider the following proposition: artificial intelligence and prediction markets are made for each other. Major conundrums facing humanity are essentially prediction problems. Will the climate change? Will the economy grow? Will the next pandemic be worse than the last one? How will the rate of climate change/economic growth/pandemic severity alter if we…
Deprecated. Just use deadlines.info
Automatic programming is a pursuit of a seemingly clear goal: let’s make computers program themselves! As usual, that nasty pal of ours is in the details: are compilers automatic programming? They do generate code automatically, and modern compilers tend to utilize sophisticated machine learning to do so [Leather, Cummins]. So, in the interest of unambiguous nomenclature, here’s a…
A traditional language model is a probability distribution $p_\text{LM}(t)$ predicting how likely text $t$ is to appear in a training corpus approximating all of language. A 2023 language model (GPT-4 [OpenAI 2023], Claude [Bai et al 2022], Alpaca [Taori et al 2023]) is a probability distribution $p_\text{LLM}(t)$ estimating the likelihood that a certain text $t$ satisfies several constraints: $t$…
An Aritificial Intelligence PhD and a Healthcare PhD walk into a bar. The Healthcare PhD starts telling the AI PhD tales from the trenches of clinical research. Tale 1: Patient parsing The UVA/PADOVA Type 1 Diabetes Simulator is an example of a mathematical smulator used for in sillico trials of diabetes treatments. The researchers assessed a dataset of diabetes patients collected in a bihormonal…
Clinical protocols Clinical protocols are key to quality health care delivery. It may come as a surprise to some, but doctors and other clinicans don’t approach every patient as a creative blank slate: every hospital has numerous clinical protocols: well-defined instructions used to solve a class of specific problems. In other words, algorithms. Now, as someone who works with algorithms in a…
In theory, Reinforcement Learning is a discipline concerned with algorithms for decision-making in opaque environments that maximize cumulative reward. In practice, however, (and yours truly is guilty of this as much as everyone) reinforcement learning literature mostly discusses solving Markov Decision Processes. However, some decision making settings are fairly hard to model as a Markdov…
What is anthropodidactic learning? Anthropodidactic machine learning is using didactic materials developed for human students (textbooks, lectures and/or lecture notes, explanations, homeworks, exercises, games and other sorts of interactive edutainment) to train artificial intelligence. Examples of anthropodidactic learning include using language textbooks to train a machine translation model or…
To quote from Wikipedia that, in turn, quotes from Technopedia a software build is “the process of converting source code files into standalone software artifact(s)”. Data Scientists often fancy themselves exempt from the art of building software artifacts (as well as a few other software development skills, but that’s a topic for another post), since they write most of their…
Let me start this essay, as one does on the internet, by polarising my audience But what if I told you that the Montecchi and Cappelletti of gradient free optimization work suprisingly well in tandem? Background: gradient-free optimization Consider the task of finding the minimum of a function without any access to it’s derivative: Here’s the function: \[ O(x) = sin x + sin 10x -…
Prescriptum: this is a tutorial on writing a custom OpenAI Gym environment that dedicates an unhealthy amount of text to selling you on the idea that you need a custom OpenAI Gym environment. If you don’t need convincing, click here. We assume decent knowledge of Python and next to no knowledge of Reinforcement Learning. Reinforcement Learning arises in contexts where an agent (a robot or a…
Before we start, let’s get some Python plumbing out of the way import numpy as np from itertools import islice # I usually program in languages where this is built in :) # https://stackoverflow.com/questions/6822725/rolling-or-sliding-window-iterator-in-python def window(seq, n=2): "Returns a sliding window (of width n) over data from the iterable" " s -> (s0,s1,...s[n-1]), (s1,s2,...,sn),…
Introduction and related work Information funnel model of neural networks Although neural networks have proven to be invaluable in research as well as in automating various business processes, a lot of questions remain open as to how they work and why they are so effective. One model developed (by Shwartz Ziv, Tishby) to answer them is the information plane From a statistical point of view, the…