Back to AI Story

Supervised and Unsupervised Learning

Navigating the Complexities of Training Data

Straightforward Human-Created Data

For certain applications like self-driving cars, the training data can be straightforwardly and precisely curated by humans. All it requires is a human touch—someone to drive.

When Data Collection Becomes a Challenge.

Consider the daunting task of predicting earthquakes. The input data here comprises geological information, while the output signifies the occurrence or non-occurrence of earthquakes. But how much data is truly enough? Unlike driving a car, we can't artificially induce an earthquake to gather more data.

Similarly, crafting a model to detect hacker intrusions presents its unique challenges. With hackers continuously devising novel methods of attack, how do we amass relevant data? And does our archive of past attack data lose relevance in the face of ever-evolving cyber threats?

Crafting the Test: The Challenge of Information Depth

Even if you make the answer sheet easy...

Consider a model for predicting stock prices. The test paper would be various economic information for the day, and the answer sheet would be whether the stock price went up or down the next day. The answer sheet is easy to assemble.

How much information to include on the test paper?

Designing the test paper presents its challenges. Is it enough to simply consider metrics like interest rates, price indexes, and unemployment rates? With events like the COVID-19 pandemic profoundly impacting the economy, isn't it imperative to incorporate such data? The outcomes of elections also sway the economy, suggesting the need to factor in news information. Moreover, if extended rainy seasons can influence the economy, shouldn't weather data be considered?

In reality, determining the perfect breadth and depth of information for the test is a multifaceted challenge.

The Domain of Supervised Learning in AI

When Test Papers and Answer Sheets Abound...

A significant portion of recent advancements in AI comes from scenarios where humans can proficiently design both the test and its corresponding answers. This paradigm is termed "supervised learning." After being trained on this human-curated data, AI is equipped to undertake tasks on our behalf.

Supervised Learning: A Human-led AI Evolution

Under the guidance of vast datasets, AI can achieve outcomes that mirror or even surpass human speed and accuracy. With continued human instruction, AI refines its capabilities, gradually assuming roles once reserved for humans.

The Landscape of Unsupervised Learning in AI

Expectations from an AI Without Human Tutoring...

Certain realms—like disaster prediction, cybersecurity, and stock markets—resist supervised learning. These are domains where human proficiency is either limited or inherently uncertain. Herein lies the scope for unsupervised learning. In the absence of substantial human-guided data, what caliber of results can we anticipate from AI?

Unsupervised Learning: AI Tackling the Uncharted

The ambition of unsupervised learning is to navigate challenges that typically confound human understanding. However, the outcomes generated by AI through unsupervised methods still necessitate the discernment and judgment of human intellect.

Back to AI Story