Skip to content

AI

essentially, from a set of information A, try to predict the value of B

eg: given a license plate, find car plate number

AI Overview

What is AI?

Artificial Intelligence (AI) refers to systems and models designed to perform tasks that typically require human intelligence, such as recognizing patterns, making predictions, and automating decisions.

Types of AI Training Models

  • Supervised Learning
    • Maps inputs (A) to outputs (B).
    • Requires labeled data (input-output pairs).
    • Model performance improves with larger and higher-quality training datasets.
  • Unsupervised Learning
    • Identifies patterns or structures in data without explicit labels.

Data in AI

How to Acquire Data

  • Manual labeling.
  • Observing existing behaviors and patterns (behavioral data).
  • Downloading datasets from online sources.

How to Use Data

  • Start with a small, high-quality dataset.
  • Train AI and analyze results.
  • Iteratively adjust data collection based on model feedback.
  • Avoid dumping random or massive uncurated datasets onto AI teams — only valuable, relevant data matters.

Common Data Problems

  • Empty entries (missing values).
  • Incorrect labels.
  • Inconsistent formats (e.g., mixed MIME types, unstructured data).

Machine Learning (ML) vs. Data Science (DS)

Aspect Machine Learning (ML) Data Science (DS)
Focus Generate predictions based on data Analyze data to arrive at insights and conclusions

Deep Learning and Neural Networks

  • Takes multiple inputs.
  • Passes data through several hidden layers of "neurons."
  • Processes and transforms information layer-by-layer to output a prediction or decision (B).

AI Companies Best Practices

  • Short Iteration Time: Rapid testing and feedback loops.
  • A/B Testing: Constant experimentation and validation.
  • Decentralized Decision-Making: Empower engineers and specialists to make key decisions.
  • Strategic Data Acquisition: Focus on collecting the right data.
  • Unified Data Systems: Ensure data is standardized and integrated.
  • Automation: Wherever possible, automate repetitive tasks and pipelines.

What AI Can (Currently) Do

  • Tasks that a human can typically complete within 1 second (simple, intuitive tasks).
  • Handle large volumes of data efficiently.

TODO:

  • Add diagrams explaining supervised vs. unsupervised learning.
  • Add a sample neural network architecture image.