Conversation with an expert on Machine Learning ( ML)
Nothing Compares to a Beautiful Conversation with a Beautiful Mind.
Last week, I had a chance to interact with one of my new friends (:P) and the hard core technologist. Because of my deepest interest in the principles of machine learning, I began to ask her few personal interests and technical questions about machine learning technology. I found her responses very insightful and encouraging which is why I decided to share the same with you.
— — — — — — — — — — — — — — — — — — — — — — — — — —
Before posting specifics of our chat, let me introduce her to all of you. Sakshi Agarwal — an ML Engineer at Arche Information Inc. in Tokyo, Japan. She did her masters from JAIST, Japan where she studied improving educational games using ML Data Analysis. In her free time, She likes to take photos. When it comes to hobbies, She has like zillions of them. She likes scrap-booking, calligraphy, learning languages, etc. Getting anything done is one of her superpowers only as long as She get an unlimited supply of coffee. If not, She can sleep for days straight :P
Me: Name the technologies are you working on?
Sakshi: I work on a varied array of items. On my normal days I use Machine Learning to focus on Image Recognition. I used different libraries, including TensorFlow, PyTorch, Chainer, etc. And then I find myself creating a 3-D world every once and a while, using three.js, Blazor, and C #.
Me: What does Machine Learning as a Technology mean to you?
Sakshi: Arthur Samuel described Machine Learning as the, “It is a field of study that gives the ability to the computers to self-learn without being explicitly programmed”. ML is mainly focused on the development of computer programs that can teach themselves to grow and change when exposed to the new data. ML is important today because the data has been growing every day and it is going to be nearly impossible for a human to understand it all. With computers, we get higher accuracy and much faster speed.
Me: What are the different types of Machine Learning?
Sakshi: In general, any machine learning problem can be assigned to one of two broad classifications:
- Supervised Learning — In supervised learning, we are given a data set and already know what our correct output should look like, having the idea that there is a relationship between the input and the output. One common example is Spam Filtering. We tell the machine if the email is spam or not. Then once we give the machine enough examples of spam and non-spam emails, it will learn to classify them accordingly. We see this, especially in Gmail accounts.
- Unsupervised Learning — Unsupervised learning allows us to approach problems with little or no idea what our results should look like. We can derive structure from data where we don’t necessarily know the effect of the variables. We can do so by clustering. For example, if we were to define clothes and shoes, the machine will make clusters with similar items. As a result, all the clothes will be in one cluster and all the shoes will be in another. With unsupervised learning, there is no feedback based on the prediction results.
Then there is a third kind which exists somewhere in between. It is called Reinforcement Learning — It is like training any animal. When a target task is fulfilled, you give a treat. The famous Alpha-Go is trained this way. Here is a famous clip of the game Atari by Google.
Me: How is ML different or related to AI and deep learning?
Sakshi: AI is a science like mathematics or chemistry. It studies the ways for machines to solve problems and build intelligent programs that have always been considered as a privilege for humans.
ML is a subset of AI which allows the system to learn automatically and improve from the experience without being explicitly programmed.
Deep Learning or Deep Neural Learning is a subset of ML which uses neural networks to analyze different factors in a system that is similar to the human neural system.
Sharing one picture with you, maybe this picture explains it better:
Me: How difficult or easy it is to work on Machine Learning? (My Fav. Ask)
Sakshi: Working on ML is not very difficult. Let’s say there are three components in ML:
According to me, if you take good enough time on the first two, having the right algorithm won’t be that difficult. Let’s take the usual example. Suppose we want to predict the price of a house. It is a supervised problem. For this, we need to collect a lot of data about all kinds of houses and in different places. If we don’t do that, our datasets could be biased and not give accurate results. Then, we need to find what features we want to use to predict the price. For example, color has nothing to do with the price, so if we use color to predict prices, we might be able to live on another planet but we aren’t going to be able to predict the prices. So, we need to use the area of the house and in some cases even the locality of the houses. If we do these well, finding the right algorithm is usually a try and hit kind of thing. It is possible to solve the same problem with different algorithms. By the right algorithm, it is implied that the algorithm must be accurate and faster to achieve the problem. [Very well explained]
There aren’t any easy or hard ML models. It is about the problem that you are trying to solve and the approach is taken to achieve it.
Me: What inspired you to choose ML and what are the core skills required to get started with Machine Learning?
Sakshi: ML is such a vast field of study currently and so many things to learn. There is so much we can do with ML, especially, automating redundant tasks. Also, thinking of it in a way where we can apply ML to almost any field. One day you could be automating things for a Motor company and the second day you could be doing it for a paper printing company. There are endless possibilities.
Core skills required for ML are Problem Solving Skills, Basic Statistics, Python, and rest you can learn as you go.
Me: What do you understand by Image Classification using ML? What are the typical use cases of image classification in ML?
Sakshi: Suppose we have 50 images. We want to know how many of them contain a cat and how many of them contain a dog. This is a typical image classification problem. It is used for Face Recognition, in self-driving cars to detect the objects around the car, check whether the product a company is producing is good or not good, etc.
Me: Can you please elaborate on the future in ML/AI/IoT?
Sakshi: Although AI will displace some jobs and such displacements have always occurred even before AI was in the talks. We have seen the reducing jobs for travel agents, milkman, elevator operator, etc. and at the same time jobs like App Developers, Data Scientists, etc.
ML has a lot of prospects in many fields. A couple of years ago, Siri and Alexa were something impossible and now it is a household name.
ML is already applied in various fields covering everything from business to scientific research:
Logistics and supply chain
Software security systems
Medicine and healthcare
Physical objects’ security systems
Big Data processing and predicting
Customer interaction personalization
Here are some areas of activity it will develop most intensively shortly soon:
1. Big Data Processing — Big data processing solutions find application in fintech (an application based on machine learning automates transaction processing) and network security (it aids identifying both existing viruses/worms/attacks and the generally anomalous behavior).
2. Advanced Search Engines - Netflix streaming service uses ML in its search engine to offer users customized content. This technology is used in the development of many upcoming e-commerce products, simplifying the interaction of users with the interface and helping to urge them to the final stage of the sales funnel. You can do the same to optimize the user experience. You can also add the functions of “smart” voice search and image search.
3. Forecasting and Analytics — Even with specialized applications, financial analysts spend a lot of time generating reports. Moreover, their work is prone to the influence of the human factor (errors in analytics incur losses in thousands of dollars). That’s why it is much more convenient to load data into special ML-based software and get the most accurate forecasts after only a few seconds, rather than being guided by the subjective vision of an experienced analyst.
4. AI-Based Software Solutions — Machine learning is one of the subsets of artificial intelligence and it is often used in the development of AI projects. Artificial intelligence becomes one of the most promising IT activities and if you want to create a successful product, you should look in the direction of artificial intelligence solutions. These are products for creating a personalized user experience — be it a smart assistant or a marketplace’s search engine.
5. Educational Applications -Today, people yearn to study remotely to save time and to choose the courses that they like. Adaptive individualized learning is one of those industries that would benefit the most from the future scope of machine learning. ML will not only help to plan individual schedules and to choose the optimal pace but also allow talented teachers to reach much larger audiences as fast as possible.
6. Biometrics Authentication - Modern smartphone users already cannot imagine their lives without fingerprint authentication. In the nearest future, it is going to enter every high-security sphere. For example, some countries already introduce biometric citizen IDs. The competition in this niche is not too tough yet. So, there is a place for innovative products.
And with that, I began asking her about higher research experience and experiences abroad. I always enjoyed talking to her and learned about Machine Learning core concepts, of course. Hope you will like it too!!
Please feel free to write at firstname.lastname@example.org for any queries and stay tuned for future talk with different technologies experts.