A listing of my articles in deep learning reinforcement learning & computing
Welcome to my articles on Deep Learning, Reinforcement Learning, and computing in general. The purpose of this article is to give a road map of the articles I wrote so far. This will create a more organized index. The first part contains the major series I wrote and the second part contains a listing of articles. At the end of this article, I will answer some questions regarding the basic guideline, credits, and policy for this blog.
This is one of the biggest Series that I want to tackle next: Deep Reinforcement Learning. This is an outline of the…medium.com
This series covers the basics of quantum computing. The first 2 articles cover the basics followed by how it is built…medium.com
Are we there yet? In this GAN series, we examine how GAN is applied to deep learning problems and look into its…medium.com
This article studies the motivation and the direction of the GAN research in improving GANs. By reviewing them in a…medium.com
In my previous article, we look into the technical details on how AlphaGo beats the Go champion. Even AlphaGo is…medium.com
How does reinforcement learning join force with deep learning to beat the Go master? Since it sounds implausible, the…medium.com
In this series, we will take a comprehensive journey on object detection. In Part 1 here, we cover the region based…medium.com
In part 2, we will have a comprehensive review of single shot object detectors including SSD and YOLO (YOLOv2 and…medium.com
Detectors, like region-based detectors or single shot detectors, start from different paths but look much similar now…medium.com
Detecting objects in different scales is challenging in particular for small objects. We can use a pyramid of the same…medium.com
You only look once (YOLO) is an object detection system targeted for real-time processing. We will introduce YOLO…medium.com
It is very hard to have a fair comparison among different object detectors. There is no straight answer on which model…medium.com
In a previous article, we discuss the use of region based object detector like Faster R-CNN to detect objects. Instead…medium.com
Three major components of autonomous driving are localization (where am I), perception (what is around me), and control…medium.com
In the previous article, we discuss how Kalman Filter works. The mathematical solution for Kalman Filter assumes states…medium.com
Three major components of autonomous driving are localization (where am I), perception (what is around me), and control…medium.com
GPS alone is not accurate enough for self-driving car. According to US government, GPS-enabled smartphones have an…medium.com
Path planning is about finding a safe, comfortable and efficient path to maneuver around the traffic. The following…medium.com
Fabrication of celebrity porn pics is nothing new. However, in late 2017, a user on Reddit named Deepfakes started…medium.com
Memory is a crucial part of the brain and the computer. In some areas of deep learning, we extend the capabilities of…medium.com
This article covers the shortcomings on the learning efficiency of current deep networks. We will look into different…medium.com
In a 6-part series, I will explain the whole journey from starting to finishing a deep learning (DL) project. We will…medium.com
The success of a Deep Learning project depends on the quality of your dataset. In the 2nd part of this series, we…medium.com
In part 3, we cover some high level deep learning strategy. Then we go into the details of the most common design…medium.com
In troubleshooting a deep network, people jump to conclusions too early too fast. Before learning how to troubleshoot…medium.com
You have build a deep network (DN) but the predictions are garbage. How are you going to troubleshoot the problem? In…medium.com
Once we have our models debugged, we can focus on the model capacity and the tuning. In this section, we will discuss…medium.com
Brainstorming, mockups and prototyping take up a large portion of time in product designs. It will be ideal if we can…medium.com
mAP is the metric to measure the accuracy of object detectors. It is the average precision for different recall values…medium.com
AlphaGo Zero uses MCTS to select the next move in a Go game.medium.com
To create something from nothing is one of the greatest feelings, … It’s heaven.medium.com
In this section, we will sample some of the GAN’s applications and display some of their results.medium.com
It is easier to recognize a Monet’s painting than drawing one. Generative models (creating data) are considered much…medium.com
GAN models can suffer badly in the following areas comparing to other deep networks.medium.com
Training GAN is known to be hard. We regularly deal with the performance, model stability, model convergence and mode…medium.com
We work hard to produce mathematical models for deep learning. But often, we are not successful and fall back to the…medium.com
GAN is a minimax game in which opponents take action to undercut the other. This makes GAN much harder to converge, if…medium.com
How can we train GAN faster with better image quality? We have many new cost functions already. But Relativistic GAN…medium.com
In discriminative models, like classification, we may handcraft features to make models to perform better. This…medium.com
The GAN discriminator distinguishes real images from the generated one. It usually performs reasonably well. But as a…medium.com
Let’s say we collect three set of pictures: one for real pictures, one for Monet paintings and the last one for Van…medium.com
Super-resolution GAN applies a deep network in combination with an adversary network to produce higher resolution…medium.com
Stacked Generative Adversarial Networks (SGAN) composes ofmedium.com
The progressive growing of GANs trains the GAN network in multiple phases. In the first phase, it takes in a latent…medium.com
Replace all maxpooling with convolutional stride.medium.com
Mode collapse is one major concern in training GAN. In the figure below, we use an BEGAN to generate faces. For each…medium.com
A 2017 Google Brain paper “Are GANs Created Equal?” claims thatmedium.com
We have mentioned optimizing GAN is optimizing JS-divergence. This is not obvious from the cost function:medium.com
Linear algebra is essential in Machine Learning (ML) and Deep Learning (DL). Even we struggle in creating precise…medium.com
Eigenvalue and eigenvector are probably one of the most important concepts in linear algebra. It is like bumping into a…medium.com
In literature, we often deal with statements like “since the matrix is symmetric, it has orthonormal eigenvectors”. It…medium.com
In machine learning (ML), one of the most important linear algebra concepts is the singular value decomposition (SVD)…medium.com
Credits & Guidelines
AI is coming. In some perspective, it sounds like “winter is coming” in the Game of Thrones. A lot of people talk about it but we barely realize its meaning yet (at least not until the last season of GOT). Making technologies simple and easy is the first step to proliferate them. We need to connect people to the technologies beyond meaningless buzz words in the TV ads. Many people have reached out to me for the ideas they have. Majority of them are amazing. I wish I have the bandwidth to follow up the ideas in more details.
To your surprise, my blog was not intended to discuss AI. It is a playground for me to think through some Deep learning or AI problems I have. Along the way, I may explain somethings that may trouble someone. I am glad that some people find them helpful. While I rarely (close to never) promote my articles, the readership increases so significantly that some extra cares may be warranted. Balancing between substance and reader engagement is not easy. I personally feel providing many external links throughout an article is very distractive and counterproductive in showcasing the research. I try not to explain technology as a historian. When everything pops, nothing pops. I suggest readers paying special attention to the reference section for the source of information and the original research papers. The links in the external figures act as another mechanism to trace the source of information. This approach is less thorough than typical research papers. But I find that strikes a better balance for the broader target audiences and I will review this again as needed. But if you are a researcher in the specific topic, I strongly refer you back to the original research paper for a more thorough reference.
Lady Gaga said social media is like the toilet of the internet. So allows me to do some plumbing works for a second. In average, I ban about 3 people per year from my blog. This is probably about 0.0005% of the readers. When Lady Gaga said social media is like the toilet, she probably thinks about a small group of people that getting un-proportional attention.
Let me quote the CEO of Goldman Sachs, David Solomon, directly.
The other thing I’d point to that’s so important is there is a real emphasis when people are interviewing around academics and I.Q. I think it’s way overweighted,” Solomon told the students. “There should be equal emphasis on E.Q. and how you interact with people, how you relate to people, and how you connect with people.
And even in a world where more and more machines and technology are disrupting the way we do business and the way we all connect, a lot of the value is still created through human interaction and the ability to motivate people or get people to move in a direction with you. All of those are things in my experience here that had a meaningful impact on me and helped me along the way.
I resonate a lot with this comment. From the hiring to the development process in the high-tech, we emphasize a lot on intellectual characters. However, many arguments we put forward are strongly influenced by our own experiences. Without proper exploration and validation, we just get stuck in opinionated discussions. In product design, the skill of empathy is the key to user-focus products. In some perspectives, we have an issue of empathy rather than the lack of intellectual capability.
I always appreciate people pointing out errors in the articles. Some (or many) of them are written in a very short time. Without a peer review, the errors in an article are higher than an academic paper. But sometimes, without knowing your background first, it is hard to recognize whether it is a misunderstanding or a mistake. If you can state the error and the suggested changes, it will help me to figure it out better. I often have problems when people try to write the equations in plain text. My suggestion is to save the original equation image in the article and reposit it in the comment. Or write the equation in a paper and take a picture. You can just drag and drop an image into your comment. However, a very small minority (a single digit of readers) may become very judgemental, authoritarian or having temperament issue. Don’t treat mistakes like the end of the “technology” world. Good engineers and researchers should verify information from different perspectives. If the articles can lower the barrier of technology, I am happy. But please play due diligence when you need the information for a more serious purpose.
Problems can turn ugly when you are not showing your face in the virtual world. Fortunately, this is an extreme minority for now. But it does take up time to handle them. However, I do want to discourage behaviors that may not be appropriate. My intention is to bring up some awareness here rather than being a supercop here. So enjoy your readings.
I do my best in responding to comments but less with private notes. One of the common mistakes is asking questions too broad or too vague. Some questions sound like a full project request. In reality, I take this type of questions very seriously but I find it too hard to detail the response appropriately. Also, it will be very helpful if you review the question first before sending it out. In average, I need further clarifications in about 1/3 of the questions. Last, troubleshooting is never a show-and-tell business. If you are struggling, it means you are learning. There is no short cut. Don’t expect to have your questions answered if you don’t dig around. Be comfortable to play around any open source code. By avoiding digging a hole for myself, I will be very selective in answering those questions.
As a side note, I always pulse the interests of the articles by the number of claps or the response. It helps me to determine whether further reviews are needed. Sometimes, people contact me in using or translating the materials in the blog. I really appreciate that because it helps me to understand what is important. This triggers me to review the proper credits to other resources again. If it is a translation, I often ask people to put a link for your translation back in the comment section here. In general, I am as generous as many researchers or academia. I need you to find the cures for many diseases!
Be creative and be responsible. Hopefully, you are the person that brings the next major breakthrough.