<<<<<<< HEAD Tbilisi AI Meetups ======= Tbilisi Tech Meetups

Tbilisi AI Meetup Series

AI meetups focused on Data Science and Machine Learning with a primary goal to foster knowledge-sharing and networking among local data science and machine learning practitioners and enthusiasts. We aim not to shy away from the mathematical details of widely used statistical and machine learning algorithms and provide talks that the community will benefit from.

Organised by Andro Sabashvili, PhD and Shota Natenadze.

Unveiling Efficiency: Operational Research and Optimization Solvers

Date: July 27, 2024
Speaker: Levan Pavlenishvili

The talk explored the world of Operational Research (OR), the use of mathematical models to optimize complex decision-making across various fields. Levan covered optimization solvers, both commercial (such as Gurobi, Knitro, FICO Xpress) and open-source (such as GLPK, HiGHS, COIN-OR). To illustrate the applications of optimization, he also introduced a technique called Data Envelopment Analysis (DEA), which evaluates efficiency by comparing similar entities using real-world data. This tool can be applied in numerous scenarios where the efficiency of a decision-making unit (DMU) needs to be assessed.

Adaptation of Open-Source LLMs on Custom Domain

Date: July 10, 2024
Speaker: Malkhaz Suladze

In the third AI Meetup, Malkhaz Suladze, Senior Data Scientist at EPAM, discussed adapting open-source LLMs for the Georgian language through tokenizer transfer and continual pretraining. He covered tokenizers, the workings of Byte Pair Encoding (BPE), and its advantages over other subword tokenizers, focusing on Hugging Face Tokenizers. Suladze also discussed BPE tokenizer expansion with Rust pseudo code, the challenges of tokenizer transfer, and the feasibility of achieving enterprise-grade quality and inference speed for Georgian LLMs.

What if....?: Estimating Intervention Effect via Causal Inference

Date: June 15, 2024
Speaker: Andro Sabashvili, PhD

Factual questions about how to answer 'What if' questions and the motivation behind them were addressed by Andro Sabashvili, PhD, who reviewed the causal inference methods that go beyond conventional correlation AIniques, including algorithms that explain how to reason about counterfactuals.

Forecasting: Benchmarks & Practice

Date: April 27, 2024
Speaker: Andro Sabashvili, PhD

Andro Sabashvili, Phd discussed time series forecasting algorithms, benchmarks, common pitfalls, and best practices in solving forecasting problems.