Andrei Khobnia

Machine Learning🧠, Natural Language Processing📖 and Information Retrieval🔍 Search Engines, Recommenders, Chat-Bots

[Andrei Khobnia @AI_Kho_]

I am developing explainability toolkit for vector search models (siamese encoders, bi-encoders, dense retrieval models). The goal is to collect and implement specialized methods of retrieval models explainability.

https://t.co/vpBpo5BuRl

#machinelearing #ml #AI #NeuralNetworks

[Andrei Khobnia @AI_Kho_]

@captaink99 @AndrewYNg @cohere @JayAlammar @SerranoAcademy By more precise re-ranking. I have article about this kind of technique “Deep Learning in Information Retrieval. Part III: Ranking” https://t.co/UQQ5gKM2AG

[Andrei Khobnia @AI_Kho_]

New type of deep learning models, Liquid Neural Networks (LNNs), has traction https://t.co/kXKIjiefnV #NN #NeuralNetworks #DeepLearning

[Andrei Khobnia @AI_Kho_]

“19% of workers may see at least 50% of their tasks impacted by GPTs”… Analysis of labor market impact of modern LLMs: https://t.co/SUm2iKZqLg #neuralnetworks #GPT #NLProc

[Andrei Khobnia @AI_Kho_]

ViperGPT answers visual questions by generating code which uses other models https://t.co/eoRHZiti9B #NLP #neuralnetworks #DeepLearning

[Andrei Khobnia @AI_Kho_]

@siddddhesh https://t.co/hPDxAUxGIC

[Andrei Khobnia @AI_Kho_]

Interesting paper about incorporating tools (including a calculator, a QA system, search engines, a translation system, and a calendar) into big language models: https://t.co/MYo2b238xr #NLProc #NeuralNetworks #DeepLearning

[Andrei Khobnia @AI_Kho_]

Great library for training models with reinforcement learning from human feedback (RLHF) using PPO (Proximal Policy Optimization) https://t.co/fder33UKVW #MachineLearning #ML #NeuralNetworks #NLProc

[Andrei Khobnia @AI_Kho_]

Great paper which shows how huge language models like GPT-3 still fail at functional competence tasks, which often require drawing on non-linguistic capacities like formal reasoning, world knowledge, situation modeling and social cognition: https://t.co/3e7z2FK9gq #NLProc #ML

[Andrei Khobnia @AI_Kho_]

44 years after John Searle had coined term "Chinese room" and 16 years after it had been described in sci-fi book "Blindsight" by Peter Watts we actually have ChatGPT... That's a milestone🙂 #NLProc #future #neuralnetworks #scifi #ChatGPT

[Andrei Khobnia @AI_Kho_]

ChatGPT Pro available for $42/mo. https://t.co/47sEvdMIMk

[Andrei Khobnia @AI_Kho_]

Great podcast with review of InstructGPT paper: https://t.co/iC1VZMikzT #gpt #nlp #ChatGPT #NeuralNetworks

[Andrei Khobnia @AI_Kho_]

Is Nvidia’s CUDA Monopoly In Machine Learning Breaking? https://t.co/WcI5HWTuUq #ML #MachineLearning #DeepLearning #GPU #PyTorch

[Andrei Khobnia @AI_Kho_]

I am writing a series of articles about applying deep learning in information retrieval. Part III:https://t.co/Fn1dQovxhz #NLP #NaturalLanguageProcessing #ML #DeepLearning #NeuralNetworks

[Andrei Khobnia @AI_Kho_]

Is reinforcement learning from human feedback a future of model training in #NLProc for tasks where it is hard to design good metric? #deeplearning #neuralnetworks #MachineLearning

[Andrei Khobnia @AI_Kho_]

I am writing a series of articles about applying deep learning in information retrieval. Part II: https://t.co/qaPdTlOOrv #NLP #NaturalLanguageProcessing #ML #DeepLearning #NeuralNetworks

[Andrei Khobnia @AI_Kho_]

I am writing a series of articles about applying deep learning in information retrieval. Part I: https://t.co/SXBURqbqaO #NLP #NaturalLanguageProcessing #ML #DeepLearning #NeuralNetworks

[Andrei Khobnia @AI_Kho_]

On PyConBY 2021 (https://t.co/p39BZpXGUf) I will talk about our experience in adopting engineering and research best practices for mahine learning projects.

[Andrei Khobnia @AI_Kho_]

On DataStart https://t.co/0w2vfjWRcw we will talk about top research results in area of deep learning for information retrieval, technical challenges we faced during development of large scale semantic search system for search indexes contain millions of documents.

[Andrei Khobnia @AI_Kho_]

On https://t.co/y6suR4UndH online conference we will talk about why it is hard to build BERT-based production system and how our team in R&D of IHS Markit in Minsk did it for domain-specific semantic search solution.