Talks
Speakers
Events
Topics
Search
Sign in
Search
Home
Talks
Speakers
Events
Topics
Leaderboard
Use
Analytics
Sign in
search talks for
⏎
Suggest modification to this talk
Title
Description
Deep learning is the most exciting recent idea in software–but it's also intimidating. If you have no previous machine learning and deep learning experience, this talk is your entry ticket to the field. We'll start from scratch, with a look at supervised learning. Then we'll see Ruby code that trains a neural network. Finally, we'll talk about deep neural networks, and explore a wonderful new concept that's changing the field of computer graphics: Generation Adversarial Networks. See how a deep learning program can invent imaginary animals! RubyKaigi 2019 https://rubykaigi.org/2019/presentations/nusco.html#apr18
Date
Summarized using AI?
If this talk's summary was generated by AI, please check this box. A "Summarized using AI" badge will be displayed in the summary tab to indicate that the summary was generated using AI.
Show "Summarized using AI" badge on summary page
Summary
Markdown supported
In the video titled "A Deep Learning Adventure," speaker Paolo "Nusco" Perrotta provides an accessible introduction to deep learning for audiences without prior experience in the field. The talk is aimed at demystifying deep learning, presenting it as a technology rather than magic, and starts with the foundational concept of supervised learning, which entails matching inputs to outputs based on provided examples. Key points discussed include: - **Supervised Learning**: The concept is introduced as a technique for training systems to recognize patterns, such as identifying animals in images by providing numerous examples of input images and their corresponding output names. - **Training and Prediction Phases**: Perrotta explains the training phase, where an algorithm learns the relationship between inputs and outputs, followed by the prediction phase, where the trained algorithm makes predictions based on new input. - **Linear Regression**: The simplicity of linear regression is illustrated by predicting pizza sales based on restaurant reservations. The prediction process involves finding a line (two parameters) that minimizes error in approximating the data. - **Multidimensional Inputs**: The discussion progresses to scenarios with multiple input variables, leading to the need for a system that handles more complex predictions involving multiple parameters. - **Classification and Neural Networks**: Perrotta introduces classifiers for binary outcomes, such as determining profitability, utilizing concepts like matrix multiplication. He elaborates on neural networks, which enhance problem-solving by enabling the modeling of curvature in data through additional layers. - **Generative Adversarial Networks (GANs)**: One of the highlights is the discussion of GANs, where two systems work against each other to create realistic images from noise. This is exemplified through an experiment with the CIFAR-10 dataset, showcasing the impressive capability of GANs in generating lifelike images. The conclusion emphasizes that despite the sophistication of deep learning, it is fundamentally grounded in mathematical principles and systematic training processes. Perrotta also raises awareness regarding biases in AI systems, highlighting the importance of scrutinizing the ethics surrounding AI applications in sensitive contexts. Overall, the session wraps up with an invitation for further questions and promotes the speaker's upcoming book, which delves deeper into these topics and is intended as a resource for those wishing to learn more about deep learning principles.
Suggest modifications
Cancel
