Natural Language Processing Bootcamp

A well-prepared learner should have significant experience with Python and entry-level experience with probability, statistics, and deep learning architectures.

Learners should also have the ability to write a class in Python and add comments to their code for others to read.

Lastly, learners should have familiarity with the term “neural networks” and the differential math that
drives backpropagation.

Learn text processing fundamentals including stemming and lemmatization. Explore machine learning methods in sentiment analysis. Build a speech tagging model.

Course Project

Part of Speech Tagging

Use several techniques, including table lookups, n-grams, and hidden Markov models, to tag parts of speech
in sentences, and compare their performance. This project demonstrates text processing techniques that
allow one to build a part of speech tagging model. Work with a simple lookup table and progressively add
more complexity to improve the model using probabilistic graphical models. Use a Python package to build
and train a tagger with a hidden Markov model, and compare the performances of all these models in a
data set of sentences

Learn advanced techniques like word embeddings, deep learning attention, and more. Build a machine translation model using recurrent neural network architectures.

Course Project

Machine Translation
Build a deep neural network that functions as part of an end-to-end machine translation pipeline. The completed pipeline will accept English text as input and return the French translation. Be able to explore several recurrent neural network architectures and compare their performance. Pre-process the data by converting text to sequence of integers. Build several deep learning models for translating the text into French. Run this models on English test to analyze their performance.

Learn voice user interface techniques that turn speech into text and vice versa. Build a speech recognition model using deep neural networks.

Course Project

Build a deep neural network that functions as part of an end-to-end automatic speech recognition (ASR pipeline. The model will convert raw audio into feature representations, which will then turn them into transcribed text. Begin by investigating a data set that will be used to train and evaluate the models.

Convert any raw audio to feature representations that are commonly used for ASR. Build neural networks that map these features to transcribed text.

Deeply understand generative AI, describing the key steps in a typical LLM-based generative AI lifecycle, from data gathering and model selection, to performance evaluation and deployment.

Laboratory 

Summarize Dialogue. 

In this lab you will do the dialogue summarization task using generative AI. You will explore how the input text affects the output of the model, and perform prompt engineering to direct it towards the task you need. By comparing zero shot, one shot, and few shot inferences, you will take the first step towards prompt engineering and see how it can enhance the generative output of Large Language Models.

– Describe in detail the transformer architecture that powers LLMs, how they’re trained, and how fine-tuning enables LLMs to be adapted to a variety of specific use cases
– Use empirical scaling laws to optimize the model’s objective function across dataset size, compute budget, and inference requirements

Laboratory

Fine-Tune a Generative AI Model for Dialogue Summarization

You will fine-tune an existing LLM from Hugging Face for enhanced dialogue summarization. You will use the FLAN-T5 model, which provides a high quality instruction tuned model and can summarize text out of the box. To improve the inferences, you will explore a full fine-tuning approach and evaluate the results with ROUGE metrics. Then you will perform Parameter Efficient Fine-Tuning (PEFT), evaluate the resulting model and see that the benefits of PEFT outweigh the slightly-lower performance metrics.

Apply state-of-the art training, tuning, inference, tools, and deployment methods to maximize the performance of models within the specific constraints of your project.

Laboratory

Fine-Tune FLAN-T5 with Reinforcement Learning

You will fine-tune a FLAN-T5 model to generate less toxic content with Meta AI’s hate speech reward model. The reward model is a binary classifier that predicts either “not hate” or “hate” for the given text. You will use Proximal Policy Optimization (PPO) to fine-tune and reduce the model’s toxicity.