Introduction
Over the past few years the term “deep learning” has become widely used in the machine learning and artificial intelligence communities. But what actually is deep learning? And where does it fall into the categories of Artificial intelligence and machine learning? Well, that’s what the article What Is Deep Learning? by Ben Dickson on Medium aims to explain more(the article can be found here).
Summary of Article
The article revolves around a new computer program that researchers at MIT created. The computer program detected breast cancer up to five years in advance which presents a massive improvement compared to previous algorithms like this as detailed in the article. The article starts out by giving some background on machine learning and AI. It then gets into defining deep learning and the article points out that machine learning is a branch in AI and deep learning is a branch in machine learning. Based on this information I created a chart to help visualize this information which can be found below:
While showing where deep learning ranks in the overarching umbrella of artificial intelligence, the article also explains some pros of deep learning and how deep learning and neural networks enabled the project at MIT not to take years longer and not use a lot more resources. The article then gets into the fact that neural networks are an integral part of deep learning and then details how they work. the article explains that each layer in a neural network will detect a feature and the deeper the layer the more nuance the feature it detects. All of these observations come together to produce a result for the end user. There are many real-world applications of deep learning and neural networks that the article details such as voice and speech recognition and natural language processing. After, the article touches on the limits of deep learning which it outlines to be the following:
- Data dependency
- Explainability
- Algorithmic bias
- Lack of generalization
Finally, the article highlights what it thinks the future of deep learning will be the continuous improvement of its algorithms and for the algorithms to overcome the pitfalls that they face now.
My Take
I really liked this article as it highlighted a lot of the key aspects of deep learning. the article also explained the concepts in a very user-friendly way. One major way that the article did this was through the use of visualizations and one very good visualization that was presented in the article is the one involving layers and how each of them helps in the overall output of the algorithm. This particular visualization showed me the concept of how neural networks’ layers progressively become more nuanced in the things that they look for. I do have one question, however, how do the layers know what each of them is looking for? Can you code that? Is it just figured out by the algorithm on its own? How do they know? Another thing that I really liked about this article was the hyperlinks to everything. If you need more information about a certain topic chances are there is a hyperlink on that topic in the article that you can use. These are a few of my thoughts on this article.
Conclusion
Overall, I really liked the article, it encompassed the major parts of a neural network, explained them in a very user-friendly way, had good visualizations to visualize the points in the article, and linked to extra resources if the user wanted/needed them. I recommend you go check it out when you get the chance(the article can be found here)!
Additional sources:
- What Is Artificial Intelligence (AI) ? | IBM. www.ibm.com/topics/artificial-intelligence#:~:text=At%20its%20simplest%20form%2C%20artificial,in%20conjunction%20with%20artificial%20intelligence.
- What Is Machine Learning? | IBM. www.ibm.com/topics/machine-learning#:~:text=the%20next%20step-,What%20is%20machine%20learning%3F,rich%20history%20with%20machine%20learning.
- “What Is Deep Learning? – Deep Learning Explained – AWS.” Amazon Web Services, Inc., aws.amazon.com/what-is/deep-learning/#:~:text=Deep%20learning%20is%20a%20method,produce%20accurate%20insights%20and%20predictions.