Why Deep Representations? (C1W4L04)  Summary and Q&A
TL;DR
Deep neural networks work well because they are able to detect simpler features in earlier layers and then combine them to detect more complex objects in later layers.
Key Insights
 😒 Deep neural networks are able to detect simpler features in earlier layers and use them to recognize more complex objects in later layers.
 💦 The hierarchical representation in deep learning mirrors the way the human brain is believed to work.
 🥘 Deep networks can compute certain functions more efficiently and with fewer hidden units compared to shallow networks.
 🍉 The term "deep learning" has helped popularize neural networks with many hidden layers, but the effectiveness of a network depends on the specific problem.
 😥 Logistic regression can be a good starting point before gradually increasing the number of hidden layers for better performance.
 🗯️ Deep neural networks have shown promising results for various applications, but it is important to find the right architecture for each problem.
 😯 Deep learning has had a significant impact in fields like face recognition and speech recognition.
Transcript
we've all been hearing that deep neural networks work really well for a lot of problems it's not just that they need to be big neural networks is that specifically they need to be deep or to have a lot of hidden layers so why is that let's go for a couple examples and try to gain some intuition for why deep networks might work well so first what is... Read More
Questions & Answers
Q: How does a deep neural network differ from a shallow one?
Deep neural networks have multiple hidden layers that allow for the detection of complex objects by combining simpler features. Shallow networks with only one hidden layer would require exponentially more hidden units to compute the same function.
Q: What is the intuition behind deep neural networks?
Deep neural networks are inspired by the hierarchical representation of the human brain. Both start by detecting simple features and then build upon them to recognize more complex objects.
Q: Why has the term "deep learning" become popular?
The term "deep learning" has gained popularity as it captures the idea of neural networks with many hidden layers. It has helped to brand and market the concept, even though networks with fewer layers can also be effective.
Q: Should deep neural networks always have many hidden layers?
No, the number of hidden layers depends on the problem. Starting with logistic regression and gradually increasing the number of hidden layers can be a good approach to find the optimal architecture for a neural network.
Summary & Key Takeaways

Deep neural networks are effective for problems like face recognition and speech recognition because they can detect simple features like edges or basic audio waveforms in earlier layers.

The earlier layers of a neural network detect simple functions, while the deeper layers combine these functions to learn and recognize more complex objects.

Deep neural networks can detect faces, words, phrases, or sentences by building upon simpler features and composing them together.