morningdecember_facebook

 

Dear Morning@Lohikan`s !

Thanks for participating and, besides sharing your knowledge, being so active in donating for IT Mykolay. We got  10 530 UAH!

Also, let’s refresh our memory and look back on what was going on last Saturday 3rd of December.

We had two great speakers and themes:

oles_petriv

Oles Petriv (Machine Learning Engineer at VideoGorillas (videogorillas.com)


Working with natural language processing, computer vision systems (realtime video analysis, video captioning), object detection systems. Interested in words semantic space analysis, generative visual models and data visualization.

Theme: Dynamic Concept Space Analysis using Machine Learning
Goal: Create clear concept space model for natural english (or other) language which could represent concepts similarities and track changes in meaning of concepts in long time periods (years and tens of years).
• creating vector representations of words using different word embeding models (word2vec, GloVe, word2gauss, fasttext)
• combining vector representations of different models to improve semantic simmilarity quality
• creating datasets of natural english language splited to timeframes (1 dataset per year)
• creating models for each year
• finding clusters of words
• tracking movement of clusters and words trajectories
• checking of ‘innertion hypothesis’ on words trajectories
• training predictive neural net models to predict future word`s positions in vector space
• validating results
• creating dimention reduction to 3 for every timeframe
• creating visualization of words movements for better understanding

ivan-lavriv

Ivan Lavriv (Web engineer Lohika)

Web branch leader at pllug.org.ua and coach for kottans.org, Drummer

Theme: Exploring machine learning in JavaScript
Discussed how and where to use JavaScript to machine learning solutions, And explore existing JS toolkits and libraries for Neural network and machine learning.
Demo description: We solved Kaggle Titanic in JavaScript and send pre – trained model as stream to browser.

Please, leave feedback below!

Powered byTypeform

Leave a Reply

Your email address will not be published. Required fields are marked *