Basic scenarios and methods — A time series is a sequence of numbers over time. Such as daily closing prices of a certain stock, or daily high temperatures in a particular location. Anomalies in time series are values that are highly inconsistent with those expected around at those time points. Anomaly detection has numerous use…

Designed for rapid assimilation — This post is designed for college students who need to understand or review significant graph algorithms. Algorithmic problems and algorithms are illustrated with Figures. (Readers unfamiliar with graphs or their basic terminology may want to read https://medium.com/p/a54ba3f25639 first.) Breadth-First Search (BFS) This problem inputs a graph G and a particular…

In an interactive, content+quiz format — This covers basic concepts of clustering and hierarchical clustering followed by some algorithms: K-means clustering, DBScan, and bottom-up hierarchical clustering. Paper and pencil may help. Quiz-style questions are followed by answers downstream. Clustering is the process of grouping similar data points together, keeping dissimilar ones in different clusters.

I asked some questions on this topic to both Bard and ChatGPT, starting with an overview question and then drilling down. I have included verbatim the responses of both chatbots to each question. I have highlighted, in bold, some text in each chatbot’s response that I felt was worth highlighting. …

From a developer’s perspective — In [1], we covered the fill-in-the-blanks self-supervised approach to learning a language model in some detail. We also presented a statistical algorithm for it that is straightforward to implement from scratch. This post is a companion to that post. From a developer’s perspective. It will be of interest to readers…

Statistical bidirectional modeling and inference — This post will cover a fill-in-the-blanks approach to training a language model. We will explain why this is a powerful self-supervision task that can be used to train a language model that is well-suited for answering questions and generating new content. We will also describe an algorithm for it that…

Why it's powerful and how to solve it — Predicting the next word has a long and successful history in language modeling, most recently in large language models. It leverages the existence of huge corpora of text: Wikipedia, public web pages strewn over the globe, and others. It is more powerful than other types of unsupervised learning on these…

Some supervised, some unsupervised, some self-supervised, in NLP and computer vision — Single-task learning is the process of learning to predict a single outcome (binary, multi-class, or continuous) from a labeled data set. By contrast, multi-task learning is the process of jointly learning to predict multiple outcomes on inputs of the same modality. Such as images or text. The obvious question to…

Using a Trie with machine-learned elements — Expressing something well takes effort. Fortunately, chatbots that embody modern NLP, such as ChatGPT, are very helpful. In this post, we will model this problem using a Trie, suitably enhanced. This Trie will automatically detect short word sequences that appear repeatedly in the corpus in an unsupervised way. …