Advanced Machine Learning and Deep Learning Projects

Explore SBERT sentence transformers introduction, using BERT-based embeddings for text and image data. See how a siamese network accelerates semantic textual similarity by turning text into embeddings for inferences.

Explore how siamese networks with triplet loss power sentence transformers, generating BERT-based sentence embeddings to measure semantic similarity and enable semantic search and retrieval.

Learn to generate sentence embeddings and compare similarity with cosine similarity using sentence transformers and pre-trained models from Hugging Face, for sentences, paragraphs, or documents.

Learn to generate sentence embeddings with a pre-trained sentence transformer, compute cosine similarity across sentences, and use paraphrase mining to identify top matching documents.

Explore semantic search with sentence transformers, embedding queries and documents into a shared vector space to capture contextual meaning, and learn symmetric versus asymmetric search models.

Learn to build semantic search pipelines with sentence transformers by embedding a corpus and queries, normalizing vectors, and using semantic_search to retrieve top matches.

Apply fast text clustering on a large Quora questions set using sentence transformers and community detection to reveal meaningful clusters from 1000 samples.

Build a multilingual Quora questions auto-complete suggester by loading a sentence transformer model, encoding 37k questions into embeddings, and performing semantic search to return top-k related questions.

Explore a similar research paper recommendation system using sentence transformers to measure semantic similarity between a paper’s title and abstract, enabling symmetric search with the Specter algorithm.

Develop a similar research paper recommendation system by loading the EMNLP dataset (2016–2018), building a title‑and‑abstract corpus, and computing sentence-transformer embeddings for semantic similarity.

Implement a deep search based recommendation system that embeds a paper title and abstract, then uses semantic similarity to surface the top three related research papers from the corpus.

Build a question-answer retrieval system using NLP by encoding documents via corpus embedding and questions via question embeddings into a shared vector space, then apply semantic search for top matches.

Learn extractive text summarization with sentence transformers, selecting key sentences to form concise paragraphs, and contrast it with abstractive methods; explore real-world examples and setup in Colab.

Apply extractive text summarization with sentence transformers and LexRank to identify top central sentences by cosine similarity, demonstrated on a Hindustan Times article.

Set up a text-to-image search using sentence transformers and clip to embed text and images in a vector space, then code a live search with 2000 images in Google Colab.

Build a text-to-image and image-to-image search pipeline by projecting text and image data into a vector space and retrieving top results via semantic search with a clip model.

Explore multilingual image search using sentence transformers and a multilingual clip model, enabling cross-language queries in Spanish, French, and more than 50 languages, via Google Colab.

Embed images with img embed and text with a sentence transformer, then use semantic search and cosine similarity to retrieve top multilingual results across Hindi, Spanish, and French.

Explore unsupervised image clustering using sentence transformers and the OpenAI clip model to generate image embeddings from the Unsplash dataset and cluster them in an embedding space.

Continue unsupervised image clustering by building a cosine similarity-based community detector over image embeddings. Apply thresholds for minimum and maximum community sizes and similarity to form clusters.

Perform unsupervised image clustering, sort extracted communities by size, keep the largest cluster first, remove overlapping and duplicates, and present clusters with three images each, yielding 26 unique communities.

Detect duplicate and near duplicate images by encoding with sentence transformers and clip, using cosine similarity and paraphrase mining to find top matches in the Unsplash dataset.