Last update:   06-09-2021

Narrow Intelligence

ANI (Narrow Inteligence) is the main AI today (like spam filters, speech recognition). AGI (General Intelligence) refers to something more complex, like emotions, gestures. AGI is just an exciting goal for researchers, but is still very far away.

Machine learning

AI is based mainly on a tool called machine learning. The core part of ML is choosing the correct algorithms.

Training data

In ML humans provide input data (training data set). Computers, using specific algorithms, learns from this data. Finally, the computers are able to output result from any input data. Deep Learning    (2/5)

Deep learning

The AI is really taken off in recent years. This is because the rise of neural networks and deep learning. Before deep learning the learning process stops at some point. This happens no matter what the amount of data.

Neural networks

With deep learning and large data sets the learning is getting better and better. Neural networks were inspired by the brain. But the details of how they work are completely different. So, Deep Learning seems to be a better name of this subset of ML.

Neural network (simple)

This is the simplest neural network with a single artificial neuron. It just inputs the price and outputs the estimated demand.

Neural network (complex)

This is an example of neural network. Its job is to learn to map fronm this four inputs. From the input A, to the output B, to demand.

Computer Vision

One of the major successes of DL has been Computer Vision
>> Image classification/Recognition (Is this a cat/you?)
>> Object detection (Is this a car/pedestrian?)
>> Image segmentation (every pixel, precise boundaries around objects)
>> Tracking (not just the position, where things are going)
Data Science    (3/5)

Data Science

Just using a simple website, users can generate data from you. This is an example of observing user behaviors. We can also observe machine behaviors, and collect data from there.

Big data

More data is better than less data. But don't just throw big data on AI team and expect miracles. It's better to consult AI team on what type of data you should collect.

Data Science

A ML project will ofthen output B from A data. In contrast, data science will extract knowledge from data.

A/B Testing

Launch two version of page and measure which button causes people to click on it. Workflows    (4/5)


Workflow of a machine learning project:
>> Collect the data
>> Train the model
>> Deploy the model
Workflow of a data science project:
>> Collect the data
>> Analyze data
>> Suggest actions

AI pipeline

Steps to process the command (example):
>> Trigger word/wakeword command (Audio -> "Hey device")
>> Speach recognition (Audio -> "Tell me a joke")
>> Intent recognition (joke?, time?, weather?)
>> Execute joke

Team roles

>> Software engineers (joke exection, tell weather)
>> Machine Learning Engineer (gathering data, A->B mapping)
>> Machine Learning Researcher (academic literature, find new tehnics)
Unsupervise Learning    (5/5)

Unsupervise learning

Supervised learning is very valuable, but it needs so much labeled data. No parent ever pointed out 10.000 mugs trying to teach their childrens. In Unsupervised Learning we give the AI system a bunch of data. AI will try to find something interesting in the data.

Reinforcement learning

Is similar to how we might train a dog to behave. We let the dog do whatever it wanted. If it behave well we will praise it. In RL a reward signall tells the AI when is doing well or poorly.
2. Machine Learning
01 . Artificial Intelligence
Last update:   06-09-2021

        A B C D E F