While the workshop schedule in Les Houches was rather tight and exhausting, I managed to squeeze in early morning runs as well as one (and only one) later morning ski session on the Kandahar run of Les Houches. Run that was quite nice due both to the quality of the snow and to the low occupancy, as French school vacations were not yet underway. Actually this Alpine week was my first week of steady running since Xmas, as my sciatica re-awoke on the very morning we were leaving for India, as I found out then during a 5am fast & furious run before heading to CDG Airport. It had been years since my sciatic nerve hurt that much and it took me up to my arrival in Chamonix to feel okay enough when running more than 4km at once. To compensate for the loss I turned through January to pool training most mornings but this proved more complicated, both timewise and locationwise, since very few nearby pools open at 7am at least once a week. And they are relatively nearby, meaning an extra 10km of extra biking at the very least each day. Henceforth a significant part of the morn having evaporated by the time I am heading to work. Still this daily training improved my swimming to some extent, mostly my endurance to get over 2km, as my speed only improved at most 10%! As I am preparing for another half-marathon at the end of March, I hope I can get back to the speed and volume reached the past Fall, tant pis for the swimming! (Unless I run to and from the pools.)
Archive for Les Houches
[H]ouch[es]!
Posted in Running, Travel with tags #ERCSyG, Behaviours Invertebrates, Chamonix, Chamonix-Mont-Blanc, ERC Synergy Grant, half-marathon, indoor swimming, Kandahar, Les Houches, MRI, Ocean, road running, running injuries, sciatic nerve, sciatica, spine, swimming pool on February 16, 2025 by xi'anwatermarking for privacy (with no durian)
Posted in Books, Statistics, Travel, University life with tags #ERCSyG, AI regulation, AI scare, durian, large language models, Les Houches, LLM, Ocean, privacy, proper scoring rule, watermarking. Nature on February 14, 2025 by xi'an
In Scalable watermarking for identifying large language model outputs, published by Dathathri, See, Ghaisas, et (many) al. in Nature of 23 October 2024, the authors propose an algorithm to (voluntarily) watermark synthetic texts to identify them as such, through a statistical test. Here are a few quotes to relate to the authors’ solution.
“LLMs generate text based on preceding context (…) given a sequence of input text x<t = x1, …, xt−1 consisting of t − 1 tokens from a vocabulary V, the LLM computes the probability distribution pLM(⋅∣x<t) of the next token xt given the preceding text x<t. To generate the full response, xt is sampled from pLM(⋅∣x<t), and the process repeats until either a maximum length is reached or an end-token is generated.
In a watermarking scheme, a sampling algorithm is an algorithm that takes as input a probability distribution p ∈ ΔV and a random seed and returns a token.
Tournament sampling selects a token from the LLM distribution that is likely to score higher under the random watermarking functions (…) Given the selection of tokens xt based on higher g-values, we expect watermarked text generally to score higher under this score than unwatermarked text (…) [It] requires g-values to decide which tokens win each match in the tournament. Intuitively, we want a function that takes a token x ∈ V, a random seed and the layer number ℓ ∈ {1, …, m}, and outputs a g-value gℓ(x, r) that is a pseudorandom sample from some probability distribution fg (the g-value distribution).”
At the Ocean privacy workshop, someone came with the question of trusting (or not) synthetic data and I remembered this article. Suggesting watermarking for said synthetic data by having providers or agents (privately) running a disclosed or registered code that delivers a watermark which provides a strong support to the synthetic data being produced likewise. I remain uncertain this is at realistic.
bused to Chamonix [jatp]
Posted in Statistics with tags Chamonix-Mont-Blanc, cross-country skiing, ERC Synergy Grant, French Alps, jatp, Les Houches, Ocean, Pays du Mont-Blanc, sunrise, Travel, winter light, workshop on February 6, 2025 by xi'an
more oceanographers in Les Houches
Posted in Books, Kids, Mountains, pictures, Running, Statistics, Travel, University life with tags #ERCSyG, Bayesian privacy, Chamonix, CHANCE, differential privacy, ERC, ERC Synergy Grant, fairness, French Alps, Les Houches, Ocean, summer school, uncertainty quantification, workshop on February 2, 2025 by xi'an
Our second internal research workshop on privacy, covered by our ERC Synergy project OCEAN is taking place in Les Houches, French Alps, this very week with 12 researchers gathering for further brain-storming on some of the themes at the core of the project, like algorithmic tools for multiple decision-making agents, along with Bayesian uncertainty quantification and Bayesian learning under constraints (scarcity, fairness, privacy). Due to the small size of the workshop, it will take place in the same local (and uncommon) hotel as last year. And where I had a genuine blackboard delivered (to be donated afterwards to École de Physique des Houches), as the place is not used to mathematicians discussing on and around a blackboard. Even though the schedule does not leave much time for skiing, I hope for a few opportunities, given that the ski cover is quite good this year!
good morning [from] Les Houches
Posted in Mountains, pictures, Running, Travel with tags Aravid, Chamonix-Mont-Blanc, downhill skiing, French Alps, Kandahar, Les Houches, morning run, Piste Noire, sunrise, super G on April 27, 2024 by xi'an
