Filtrable HNSW Part 2 - Implementation

There is a fast C++ implementation now. You can build, for example, geo-aware vector search with it. Examples included!

Tsetlin Machine

An interesting concept of another NN killer witch might be underestimated.

Filterable approximate nearest neighbors search

How to make ANN search with custom filtering? Search for closest vectors only in selected subsets.

Partially trainable embeddings

A simple tip how to control overfitting and learnability if your data is not too large.

Category Prediction

My project on mention classification. It takes text, finds people mentions and tries to describe them by their mentions.

Shrincking Fasttext

Shrinking fastText embeddings so that it fits Google Colab

Memory Augmented

Interesting approach to combine neural inference with knowledge bases.

Allennlp Seq2seq

AllenNLP - customizable PyTorch framework with proper OOP, tons of model examples and utils.

Neural QA

What types of neural Question Answering exists and how to start building your own. Step-by-step approach.

On-Disk Embeddings

How to use large fastText embeddings if you have low RAM?

Parallel preprocessing

How to prepare batch faster than GPU could process it? How to use advantage of multiple cores properly?

FastText embeddings done right

How to use PyTorch EmbeddingBag for proper OOV handling and fastText training

Loss function porn

Experimenting with different hyperparameters and what we can tell from loss curve

Neural Networks debugging

How to be sure that the source of errors is not the code but the network architecture? How to determine limits in architecture capacity?

Matching models

Ininteraction VS representation. What kinds of matching models for NLP exists and why not all of them scales well.

Matching Zoo

Matching model collection

Transfer learning in NLP

What is NLP Omniglot? What attempts for transfer learning in NLP exists?