From AlphaGo to ChatGPT, what kind of future do we foresee?

A Machine Learning competition held by Netflix in 2006. Netflix provided a $1M prize to contestants who could improve the AI 10% improvement of Netflix's recommendation system at that time.

Data provided to the contestants at the time: 100,480,507 ratings that 480,189 users gave to 17,770 movies.

In 2016, AlphaGo beat Lee Sedol in a five-game match, the first time a computer Go program has beaten a 9-dan professional without handicap. I was also watching the tournament at the time. Seeing the game from the perspective of AI experts made AI experts realize that the era of AI is coming. Then, many experts who are interested in AI are actively involved in the field of AI.

During 2022-2023, the powerful capabilities of ChatGPT have attracted the attention of the public. Entrepreneurs can now use many AI APIs to start their business, which is the result of step-by-step accumulation that started many years ago.

Human beings will rely more and more on AI, and almost all technological applications may use AI to go further.

With the advent of quantum computers, the capabilities and applications of AI in the future will be beyond imagination. Quantum computers will give AI a more powerful learning ability, and only AI has the ability to manage the new world formed by using quantum technology.

For your reference, below are some AI APIs. Automatic produce recognition Fast, accurate identification of fresh fruit and vegetables and bulk goods like nuts and confectionery. https://www.tiliter.com/ Retrieve human insights for search relevance, ecommerce, natural language processing, audio transcription, document processing and more. https://scale.com/ Image classification is the task of classifying and assigning labels to groupings of images or vectors within an image, based on certain criteria. A label can be assigned based on one or more criteria. https://levity.ai/blog/image-classification-in-ai-how-it-works

I hope that this information will help. If you need any further information, please feel free to contact me. https://discord.gg/qHj8sHrctS

My blogs: AI club, Quantum club

A Practical Definition of Machine Learning

A practical definition of machine learning is to use data to compute hypothesis g that approximates target f.

Take credit approval as an example

input x 𝑿 (customer application) output y 𝒀 (good/bad after approving credit card)

unknown pattern to be learned ⟺ target function 𝑓: 𝑿 → 𝒀 is a ideal credit approval formula

historical records in bank

data ⟺ training examples

𝐷 = {(𝒙₁, 𝒚₁),(𝒙₂, 𝒚₂), …, (𝒙ₙ,𝒚ₙ)}

hypothesis ⟺ skill with hopefully good performance 𝑔: 𝑿 → 𝒀 is a ‘learned’ formula to be used

{(𝒙ₙ,𝒚ₙ)} → Machine Learning → 𝑔

The learning model

assume 𝑔 𝐻 = { 𝒉ₖ }

i.e. approving if

𝒉₁ : annual salary > $10,000

𝒉₂ : debt > $1,000

hypothesis set 𝐻

can contain good or bad hypotheses

up to 𝐴 to pick the ‘best’ one as 𝑔

learning model = 𝐴 and 𝐻

I hope that this information will help. If you need any further information, please feel free to contact me. https://discord.gg/qHj8sHrctS

My blogs: AI club, Quantum club

How can a machine learn our preferences?

In a previous post, I mentioned Machine Learning acquiring skill with experience accumulated/computed from data.

data -> ML -> skill

Here are a few papers that try to apply Machine Learning to our life.

Abu-Mostafa, 2012

data: sales figures + client surveys

skill: give good fashion recommendations to clients

Tsanas and Xifara, 2012

data: characteristics of buildings and their energy load

skill: predict energy load of other buildings closely

Stalkamp et al., 2012

data: some traffic sign images and meanings

skill: recognize traffic signs accurately

Sadilek et al., 2013

data: Twitter data (words + location)

skill: tell food poisoning likeliness of restaurant properly

Recommender System

One hot application of ML is recommender systems. Netflix and Yahoo have held competitions related to recommender systems.

A competition held by Netflix in 2006.

100,480,507 ratings that 480,189 users gave to 17,770 movies.

A competition held by Yahoo in 2011.

252,800,275 ratings that 1,000,990 users gave to 624,961 songs.

How can machines learn our preferences?

A possible ML Solution

Pattern: 

rating ← viewer/movie factors

Learning: 

known rating

→ learned factors

→ unknown rating prediction

I hope that this information will help. If you need any further information, please feel free to contact me. https://discord.gg/qHj8sHrctS

My blogs: AI club, Quantum club

Key Essence of Machine Learning

There are 3 key essences to help us decide whether to use ML (Machine Learning).

As we explore the Key Essence of ML, let's consider the following questions.

When can Machine Learning?

Why can Machine Learning?

How can Machine Learning?

How can Machine Learning be Better?

When can Machine Learning?

What is Machine Learning?

From Learning to Machine Learning

Learning: acquiring skill with experience accumulated from observations

observations -> learning -> skill

Machine Learning: acquiring skill with experience accumulated/computed from data

data -> ML -> skill

What is skill?

skill -> improve some performance measure (e.g. prediction accuracy)

Machine Learning: improving some performance measure with experience computed from data

data -> ML -> improved performance measure

Why use Machine Learning?

For example: Tree recognition

=> Handwriting 100 rules to identify a tree?

=> Do children learn to identify a tree by memorizing the 100 rules?

=>=>  Children understand what a tree is by observing many trees by themselves.

=> In some applications, ML is an alternative route to build a complicated system.

=>=> Let the machine analyze the data by itself and learn to do certain things by itself.

Some Use Scenarios

When human cannot program the system manually

=> navigating on Mars

=>=> We cannot know in advance what the rover will encounter on Mars, and we cannot pre-write a large number of rules so that the rover will adopt a certain rule when it encounters a certain situation.

=>=> ML let the rover on Mars learn how to adapt to the environment on Mars through observation. Let the rover learn how to act on Mars by interacting with the environment.

When human cannot “define the solution” easily

=> speech/visual recognition

When needing rapid decisions that humans cannot do

=> high-frequency trading 

When needing to be user-oriented in a massive scale

=> consumer-targeted marketing

=>=> provide personalized service for each user

Key Essence of Machine Learning

Machine Learning: improving some performance measure with experience computed from data

data -> ML -> improved performance measure

Key 1. exists some “underlying pattern” to be learned

so ‘performance measure’ can be improved

Key 2. but no programmable (easy) definition

so ML is needed

Key 3. somehow there is data about the pattern

so ML has some ‘inputs’ to learn from

The three key essences help us decide whether to use ML (Machine Learning).

I hope that this information will help. If you need any further information, please feel free to contact me. https://discord.gg/qHj8sHrctS

My blogs: AI club, Quantum club