Evolución de la IA - (5/7)

Introduction to Artificial Intelligence Evolution

Overview of Autonomous Vehicles

• María Florencia introduces the topic of artificial intelligence (AI), highlighting its evolution and applications, particularly in autonomous vehicles.

• Emphasizes that AI is not limited to terrestrial vehicles; NASA also utilizes these technologies significantly.

The ImageNet Project and Computer Vision

• Discusses the 2009 ImageNet project developed by Face Faily, which addresses critical needs in computer vision, including object categorization.

• Highlights a growing demand for high-quality reference points for object categorization and established evaluation metrics.

Data Availability and Algorithm Development

• Notes the increase in data availability due to the digital era, prompting researchers to develop sophisticated algorithms for indexing multimedia data.

• Stresses the importance of providing large-scale image databases for training and testing various AI models.

Advancements in Deep Learning

IBM Watson and Visual Recognition Challenges

• Introduces IBM Watson's involvement with ImageNet challenges during the 2010s, focusing on deep learning approaches achieving new milestones.

• Describes a large-scale visual recognition challenge aimed at quantitatively assessing AI systems' ability to identify objects within images.

Competition Structure and Objectives

• Explains that participants used a vast dataset of manually labeled images (around 10 million across over 10,000 categories).

• Competitors had to label unannotated test images accurately, identifying key objects present within them.

Breakthrough Moments in AI

IBM Watson's Trivia Success

• In 2011, IBM Watson won a trivia competition against human experts, showcasing AI's capability to understand and respond effectively to complex questions.

Neural Networks and Learning from Unlabeled Data

• In 2012, researchers trained a neural network using approximately ten million unlabeled YouTube images solely for recognizing cats.

• This marked significant evidence of artificial neural networks' learning capabilities through deep learning techniques.

Generative Adversarial Networks (GAN)

Introduction of GAN Technology

• Discusses Google's introduction of Generative Adversarial Networks (GAN), where two neural networks compete against each other.

Applications of GAN

• GAN technology generates synthetic data instances that mimic original datasets under specific criteria.

Introduction to Machine Learning and Transformers

The Evolution of Neural Networks

• In 2014, the integration of machine learning concepts with game theory began, where networks function as a generator creating data from a repository and a discriminator that determines if the data is real or generated.

• By 2015, Google released TensorFlow, an open-source library that quickly became the default software for developing machine learning projects. It aims to provide an interface for expressing large-scale machine learning algorithms.

• TensorFlow presents machine learning algorithms as computational graphs, where vertices represent operations and edges signify the flow of data between these operations.

• A basic example involves drawing edges between input variables X and Y to an output node Z, which represents the result of applying a binary operation on X and Y.

Advancements in Neural Network Architectures

Introduction of Transformers

• The Transformer model was introduced in 2017 as an artificial neural network capable of understanding context by tracking relationships in sequential data like words in sentences.

• Transformers utilize mathematical techniques known as attention mechanisms to detect subtle dependencies among elements within a dataset.

Impact on Artificial Intelligence

• Described first in a 2017 paper by Google, Transformers have emerged as one of the most powerful classes of neural networks, driving significant advancements in machine learning known as Transformer AI.

• Researchers at Stanford labeled Transformers as foundational models due to their paradigm-shifting capabilities in large-scale artificial intelligence applications over recent years.

Applications and Implications

Versatile Use Cases

• The computational power of Transformers enables them to perform tasks such as real-time translation, aiding individuals with hearing impairments during meetings or classrooms.

• They assist researchers in analyzing genetic sequences (DNA), detecting trends for fraud prevention, optimizing manufacturing processes, making online recommendations, and enhancing healthcare applications.

Data Utilization Cycle