Aleksei Ustimenko
@iseethefall
Followers
43
Following
169
Media
1
Statuses
19
CEO/Founder @ simulacra AI (https://t.co/vnoffrz16B) Building next gen of quantum chemistry AI
London, UK
Joined April 2013
Congratulations to @beffjezos and the team. It took @vkleban 72 cocktails with @trevormccrt1 to become a believer in thermodynamic computers and countless hours with @toddhylton, @mjdramstead, @JasonGFox, @iseethefall to truly understand it.
0
1
8
Происходящее с математиком Азатом Мифтаховым - пытка, растянутая во времени. По сфабрикованному делу ему дали шесть лет. За отказ давать ложные показания на невиновных Мифтахову создали условия поистине невыносимые. Его тюремный сро�� - шестилетний ад, и, когда этот ад закончился,
60
451
3K
Excited to announce that our work with @LProkhorenkova and @iseethefall on uncertainty estimation for tabular data via ensembles of gradient-boosted decision trees (GBDTs) has been accepted to ICLR2021! Paper available here: https://t.co/trZfNdStWr
1
6
28
Check out this new work by our research team! If you're interested in alternatives to vector space embeddings, make sure to read the full paper.
Introducing GraphGlove: represent words as graph nodes instead of vectors, get improved quality and learned hierarchy. This work, done with Sergei Popov, Liudmila Prokhorenkova, and @lena_voita, will appear at #emnlp2020 — looking forward to discussions! https://t.co/pDCTWzfYKB
0
3
12
Introducing GraphGlove: represent words as graph nodes instead of vectors, get improved quality and learned hierarchy. This work, done with Sergei Popov, Liudmila Prokhorenkova, and @lena_voita, will appear at #emnlp2020 — looking forward to discussions! https://t.co/pDCTWzfYKB
1
1
9
Check out our new tutorial on uncertainty prediction in CatBoost and learn about different types of uncertainty – the one in data and knowledge uncertainty😺
towardsdatascience.com
Understanding why your model is uncertain and how to estimate the level of uncertainty
2
10
42
Yandex ranks fifth in the list of companies with the highest publication count at ICML 2020! The ranking by Thundermark Capital is based on the Nature Index and takes into account the affiliation of each author. Find out more: https://t.co/d8qwy1I4DF
0
2
14
In our #ICML2020 paper we propose StochasticRank — the first globally converging learning-to-rank algorithm with provable guarantees that directly optimizes any ranking quality function. Already available in @CatBoostML! ▪️ https://t.co/n2UgNCjuAz
1
7
26
[1/2] Happy to announce our work: Uncertainty in Gradient Boosting: https://t.co/7v6ioYLmQE Gradient Boosted Decision Trees can sometimes be preferable to Deep Learning. Here, we use the ensemble-structure of GBDT models to estimate of total, data and knowledge uncertainty.
1
3
15
StochasticRank in CatBoost is out!
#CatBoost 0.23 is out! This release contains many new features, including training on huge datasets, new ranking and regression modes, text features support for CPU training and more:
0
0
0
StochasticRank: Global Optimization of Scale-Free Discrete Functions https://t.co/Z2raKLGdiY
0
0
1
New #CatBoost 0.21 release is out! The main feature of the release is Stochastic Gradient Langevin Boosting mode that can improve quality for non-convex losses. Specify langevin option and tune diffusion_temperature and model_shrink_rate. Details in https://t.co/jZnPWtlpu4.
1
9
26
A big question about AI: Is it possible to be intelligent without also having an instinct for self-preservation?
90
83
234