Posts

Week 8 - BALT 4364 - Language Models

Image
       Chapter 8 takes a deep dive into Large Language Models (LLMs), one of the biggest breakthroughs in artificial intelligence. These models—like ChatGPT and DALL·E 2—are changing the way machines understand and generate human-like text and images. In this chapter, I explore how LLMs work, how they’re trained, and why they matter, along with a hands-on exercise to help me actually apply what I’m learning. LLMs are built on transformer architecture, which allows them to recognize patterns and relationships in language. That’s what makes them so powerful for tasks like answering questions, summarizing long passages, writing emails, or even creating poetry. The chapter also breaks down the two major stages of training: pre-training, where the model learns general language patterns from massive amounts of text, and fine-tuning, where it’s shaped for a specific task. Understanding this process helps me see why LLMs are so capable—and where their limitations come from. ...
Post a Comment
Read more

Week 7 - BALT 4364 - TensorFlow vs. PyTorch — What It’s About and Why It’s Usef

Image
TensorFlow vs. PyTorch — What It’s About and Why It’s Useful             When you get into deep learning, the two frameworks you hear about the most are TensorFlow and PyTorch. Both are powerful, open-source tools, but each one has its own strengths. Knowing the difference helps you decide which one fits your goals. TensorFlow was created by Google, while PyTorch was developed by Facebook. TensorFlow has a bigger ecosystem with lots of pre-trained models and tools, but PyTorch has grown fast, especially in research. Many people find PyTorch easier to learn because it works more like regular Python, making it simple to test ideas and fix errors. TensorFlow used to be harder to use, but TensorFlow 2.0 made it much more beginner-friendly. For deployment, TensorFlow is usually the better choice because it has strong tools for servers, mobile devices, and even web apps. PyTorch has deployment options too, but they’re not as mature. TensorFlow also has Tenso...
Post a Comment
Read more

Week 6 - Balt 4364 - PyTorch

Image
  In this chapter, I dove into PyTorch, a library I’ve personally found way more flexible and intuitive than a lot of the options out there. It’s strongly connected to the fast.ai community, which I highly recommend if you like hands-on learning without getting buried in unnecessary math. Their approach makes concepts feel practical and relevant, especially with all the rapid changes happening in AI right now.  PyTorch centers around tensors—basically multi-dimensional arrays similar to NumPy, but with the power of GPU acceleration. Creating and manipulating tensors is simple, and PyTorch makes operations like addition or multiplication feel natural.  To put everything into practice, I walked through a hands-on exercise using the Boston Housing dataset. I loaded the data, standardized it, built a simple neural network with one hidden layer, and trained it using MSE loss and SGD. By the end, the model could predict housing values, and it was easy to tweak things like the a...
Post a Comment
Read more

Week 5 - BALT 4364 - Understanding Natural Language Processing

Image
  Understanding Natural Language Processing      A strong area of artificial intelligence called natural language processing (NLP) makes it possible for computers to comprehend and produce human language. Three primary NLP tasks are examined in Chapter 5: language modeling, sentiment analysis, and text categorization. The main goal of text classification is to classify written content. Examples of this include classifying news items by topic or identifying spam emails. Sentiment analysis delves deeper by identifying feelings and viewpoints in text, enabling companies to assess consumer satisfaction through social media posts or reviews. The basis of programs like ChatGPT is language modeling, which forecasts the subsequent words in a phrase and facilitates tasks like text generation, speech recognition, and translation.       Additionally, the chapter offers practical exercises for creating models with TensorFlow and LSTMs using real datasets. The...
Post a Comment
Read more

Week 4 - Balt 4364 - Understanding Deep Learning and Neutral Networks

Image
Understanding Deep Learning and Neural Networks Deep learning is an effective tool which allows computers to learn from data in a manner similar to how humans learn from experience. Although it is a component of machine learning, it is unique in that it can independently identify significant patterns. Because of this, it can be used for tasks like facial recognition, speech comprehension, and language translation. Neural networks are the foundation of deep learning. These networks, which process information using layers of interconnected "neurons," are modeled after the human brain. Neural networks are used for a variety of applications. While Recurrent Neural Networks (RNNs) manage sequences like text or time-based data, Convolutional Neural Networks (CNNs) perform well with images.
Post a Comment
Read more

Week 3 - BALT 4364 - Understanding Machine Learning and Its Core Types

Image
  Understanding Machine Learning and Its Core Types One of the key components behind today's most inventive technologies, from self-driving cars to specific suggestions, is machine learning. Machine learning is a branch of artificial intelligence (AI) that focuses on developing algorithms that let computers learn from data, spot trends, and make judgments with little assistance from humans. These systems learn from experience, becoming increasingly accurate and effective over time, rather than depending on predetermined instructions.Each of the three main categories of machine learning has a distinct function. Labeled data is used in supervised learning, which means that the inputs and accurate outputs are previously known. By understanding their link, the model is able to forecast new data with accuracy. In industries like banking and healthcare, methods like logistic regression, support vector machines, and linear regression are frequently employed. Unsupervised learning, on the ...
Post a Comment
Read more

Week 2 - BALT 4364 - Using Google Colab for Hands on Python

Image
  One of the greatest resources for studying and practicing Python programming is Google Colab. Users can create and test Python code in their browser without any setup thanks to its free, cloud-based environment. For professionals, students, and novices who would rather concentrate on code than installation headaches, this makes it ideal. Colab makes it simple to upload datasets, visualize data, and work with tools like Matplotlib, NumPy, and Pandas. Additionally, it interacts with Google Drive, enabling real-time notebook collaboration, sharing, and saving. Colab also provides free GPU and TPU access to speed up computations for anyone interested in data science and machine learning. Google Colab offers a simple and engaging method to get practical Python experience, whether you're finishing a class assignment, following an online course, or testing out new concepts.
Post a Comment
Read more