Globalization made the world flat, AI drills holes into the infinitesimally small

The function of deep learning

Untangling definitions

The function of deep learning

Untangling definitions

Algorithmic
Cybernetic
Machine learning
Deep learning

The function of deep learning

Untangling definitions

Cybernetic

Multidisciplinary
Concerned with:
- Systems
- Autonomous actors
- Feedback loops
Informed and developed in parallel with:
- Nonlinear control theory
- Information theory
- Computer science

The function of deep learning

Untangling definitions

Algorithms

Sequences of instructions
Precise, finite, predictable
No The Algorithm

Algorithms** are precise, finite and predictable sequences of instructions.

A cooking recipe is an algorithm, the instructions for the assembly of Ikea furniture are algorithms.

Especially with the generalization of algorithmic feeds on social media and recommendation algorithms **The Algorithm** started to appear in singular, as an all encompassing entity that's beyond knowing and control.

There is no such thing. There isn't even a single algorithm behind something like the Facebook feed, rather a system composed of many algorithms that not only can, but are often monitored and analyzed in order to be improved or altered in response to world events and the changing priorities of the firm that deploys them.

Some of the subsystems use machine learning.

The function of deep learning

Untangling definitions

Machine learning

Self-learning algorithm is a poor metaphor
In practice weak relation to cybernetics
Statistics
Approximates algorithms
Historically more than neural networks
- now not so much

**Machine learning** is often described as a way in which computers can teach themselves to adjust or even write their own algorithms.

This is a metaphor and a poor one at that.

Some of the intuitions used by these methods can be traced back to cybernetics, but in their practical implementation their primary instruments come from statistics.

Machine learning is a field of computation focused on creating systems that approximate algorithms, when the precise dynamics of a phenomena escape the engineers or when expressing them is much too difficult.

In recent years the field of machine learning has become synonymous with neural networks, but it contains many more approaches such as genetic algorithms, Bayesian networks and decision trees.

The function of deep learning

Untangling definitions

Deep learning

Neural networks
- hidden layers of neurons.
Needs plenty of processing power
Needs Big Data
Approximates hard to design algorithms

The function of deep learning

Untangling definitions

Complex systems
Many deep neural networks
Many algorithm subsystems

The function of deep learning

Untangling definitions

AI Sometimes it's a scam

The function of deep learning

Functions and good enough approximations

The function of deep learning

Functions and good enough approximations

\[ f:\overbrace{A}^{domain} \to \underbrace{B}_{codomain}\\ \]

The function of deep learning

Functions and good enough approximations

\[ f:\mathbb{R} \to \mathbb{R} \] \[ f(x) = x \]

The function of deep learning

Functions and good enough approximations

Warm or cold?

\[ f_1:[0..255]^3 \to (-\pi..\pi] \]

\[ f_1(r,g,b) = atan2(\sqrt{3}(g-b), 2r-g-b) \]

The function of deep learning

Functions and good enough approximations

Warm or cold?

\[ f_2:(-\pi..\pi] \to \{0, 1\} \]

\[ f_2(x) = \begin{cases} 0 & \quad \text{if } \color{#00FF00}{120} \gt x \lt \color{#FF00FF}{300} \\ 1 & \quad \text{otherwise} \end{cases} \]

The function of deep learning

Functions and good enough approximations

Warm or cold?

\[ f_3:[0..255]^3 \to \{0, 1\} \]

\[ f_3(r,g,b)=f_2(f_1(r,g,b))= \begin{cases} 0 & \quad \text{if } \color{#00FF00}{120} \gt atan2(\sqrt{3}(g-b), 2r-g-b) \lt \color{#FF00FF}{300} \\ 1 & \quad \text{otherwise} \end{cases} \]

The function of deep learning

Functions and good enough approximations

Warm or cold?