AI for Suspicious Activity Monitoring

Explore Gen AI in action, generating videos from brief prompts and turning data into insights through analysis, automation, and predictive analytics for newbies with a fun, friendly tone.

Explore the end-to-end large language model workflow, from loading PDFs and text splitting to embedding and vector storage. Then harness semantic search and an LLM to generate final answers iteratively.

Trace the evolution of GPT from GPT-1’s 117 million parameters to GPT-4’s data-to-text capabilities, noting training timelines, fine-tuning, multilingual output, text-to-text vs data-to-text, and reliability differences.

Master prompt engineering by defining tasks, setting context, providing demonstrations, and using interactive, iterative prompts with style controls to shape AI outputs while avoiding bias.

Explore prompt engineering with lang chain and OpenAI to build a text classification model that detects spam vs ham, using few-shot prompts and an extraction chain evaluated with accuracy.

Learn how prompt engineering shapes outputs in generative ai for suspicious activity monitoring, covering text-based, keyword-based, rule-based, and generative prompts, plus templating, frameworks, and prompt chaining, role play, and contexts.

Access mistral large language models on Hugging Face, including the 7 billion parameter 7b v0.1 transformer. Learn login requirements, memory needs, and the pip and tokenizer setup for text generation in PyTorch.

Explore how ai hallucinations arise in generative tools like ChatGPT, exposing the black-box risk of misleading, biased outputs and real-world missteps with bogus cases and fake research.

Apply generative ai with vision to recognize items in an image, encode the image, and flag unusual or rotted food by listing all detected items with gpt-4 vision.

Discover the foundational building blocks of data science in Python, including exploratory data analysis, visualization, statistics, and machine learning, for suspicious activity monitoring.

Explore what constitutes an anomaly and how autoencoders detect anomalies by measuring reconstruction loss in unsupervised learning, using examples like abnormal pulse and fever spikes.

Train an h2o autoencoder on a 75/25 split to detect anomalies and distinguish malignant from benign tumors, using 120 neurons, tanh activation, and feature importance.

Implement a simple autoencoder in Keras to learn a 32-dimensional encoded representation and reconstruct MNIST digits using dense layers, adadelta optimizer, and binary cross-entropy.

Set up a simple autoencoder for unsupervised anomaly detection by dropping the diagnosis column, using 30 predictors, a single hidden layer with tanh activation, and a 75% training split.

Implement a basic h2o vanilla autoencoder with tan activation, train for 50 epochs on an 80/20 split, and detect anomalies via reconstruction error and a 0.1 threshold.

Learn to perform anomaly detection with variational autoencoders using the H2O deep learning estimator, training on a prepared data set, calculating reconstruction error, and flagging anomalies above a 0.001 threshold.

Learn to extract keywords using tf-idf by preprocessing text, removing stopwords, and vectorizing chunks. Apply the method to PDFs and other text data to identify top keywords with tf-idf scores.

Configure api keys and environment, load and preprocess pdf data with LangChain, build a vector store with embeddings, and query with gpt-3.5 turbo.

Explore how tf-idf ranks word importance by balancing term frequency with its corpus-wide uniqueness, using stopword removal and pre-processing to reveal key terms.