Ivan Ukhov

Ph.D., software engineer

[Ivan Ukhov @IvanUkhov]

In this article, we shall take a look at how to build #guardrails for the number of sessions and conversion rate for a growing portfolio of online stores using #BayesianStatistics. https://t.co/fBrfe3Kmpp

[Ivan Ukhov @IvanUkhov]

When the model does not fit on a single device, the common strategy is to reduce the batch size. However, smaller batches lead to noisier updates. One solution is gradient accumulation. Here is how it can be implemented in practice. #MachineLearning https://t.co/pqFO1btJTZ

[Ivan Ukhov @IvanUkhov]

In Music Transformer, the authors presented a memory efficient approach to calculating relative #PositionalEmbedding in decoder blocks. In this article, the approach is generalized to any #AttentionMechanism. #LLM #MachineLearning #Transformers https://t.co/XpPCW1kmfY

[Ivan Ukhov @IvanUkhov]

“They say that if you enjoy sausages, you shouldn’t look too carefully into how they are made, and the same is true of OpenType fonts.”—@simoncozens #OpenType #fonts https://t.co/0bya0tzQBS

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Hur resonerar man när man klassificerar ordet förment som adjektiv, inte adverb? Tack!

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Det finns en stor kvalitetsskillnad mellan det långa och det korta u-ljudet. Gäller det också å-ljudet? Tack!

[Ivan Ukhov @IvanUkhov]

@oplundgren Juste! Bra poäng, tack! Det är ju Kjellins amöbaljud.

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Slog upp ”biologi” och ”teknologi” i Folkets lexikon, och där står det att -logi uttalas som [låg'i:] i det första men som [log'i:] i det andra. Stämmer det?

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Tack! Då antar jag att det är o som i ost i båda fallen.

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Hur blir det med ordföljd när det finns flera samordnande bisatser? Tack!

1. Det gick inte, eftersom A inte kunde och B inte ville.
2. Det gick inte, eftersom A inte kunde och B ville inte.

[Berlin Bayesians ([email protected]) @BBayesians]

We'd like to thank @paulbuerkner for an amazing talk on specifying priors in a Bayesian workflow last Thursday. We were very happy that so many showed up! You can find the recording here: https://t.co/wACZFKOfX4

[Andrew Gelman et al. @StatModeling]

Progress! (cross validation for Bayesian multilevel modeling) https://t.co/RVUnzVpKZN

[Ivan Ukhov @IvanUkhov]

@Sprakradgivning Skriver man ”där uppstår alltså inte problemet” eller ”där uppstår problemet alltså inte”? Eller kvittar det? Tack!

[Andrew Gelman et al. @StatModeling]

Webinar: An introduction to Bayesian multilevel modeling with brms https://t.co/mOYAivsT4A

[Ivan Ukhov @IvanUkhov]

Uncertain about the distinction between aleatoric uncertainty and epistemic uncertainty? Here it is again, with an intuitive example by @avehtari: https://t.co/cbFSSms9Yy #Bayesian #statistics #BDA

[Ivan Ukhov @IvanUkhov]

Tired of histograms and kernel density estimators? In this article, we describe and implement a #Bayesian approach to the estimation of data-generating probability distributions by means of the #DirichletProcess. #DataScience #RStats #statistics @matvil https://t.co/RbwTdAnQIy