Recommender Engine In The AI Era
— Recommender System, AI — 5 min read
Meet Kaumil Trivedi
AI Developer @ Frendy
Ahmedabad, Gujarat, India
Kaumil Trivedi works as an AI Developer at Frendy, a socio-commerce application where people can shop products as well as earn by forming their communities.
He works on the search functionality of the application as well as building recommendation systems in the application.
Previously, Kaumil has worked in the domains of Natural Language Processing and Computer Vision, developing enterprise chatbots and performing object detection in videos.
How did you become an AI developer?
I started my journey in the field of artificial intelligence when I was pursuing my undergraduate studies back in 2016.
My 5th semester had just been over and I had always been keen on learning something interesting during semester breaks.
I came across the term “Artificial Intelligence” when I was researching voice assistants in smartphones, and got fascinated by the domain.
To learn more about the subject, alongside my regular semester courses, I enrolled in various online courses on natural language processing, machine learning.
During that time, in July 2017, I wanted to do an internship in data science to get industry experience but at that time, there were not so many opportunities.
I applied to a lot of internship opportunities at that time but was unsuccessful. At that time, the job placements had begun in my college.
I was terrified at that time because I didn’t want to be jobless after college but at the same time, I did not want to work in any field but AI.
Since I wasn’t getting an internship, I decided to get a developer job and do machine learning after work until I can get a job in the domain.
I was offered a job as a java developer at a company and my family was really happy about it.
But then I talked to my parents about the whole situation and they encouraged me to pursue AI, following which I declined the offer and continued to hone my data science skills.
Later on, in my final semester of college (December 2017), I did a full-time internship in data science which embarked on my professional journey in the field of AI. I worked on fraud detection problems and detecting objects in images.
I learned so much from that internship about the difficult tasks in solving real-world machine learning problems like data collection and data cleaning.
And then later, in June 2018, I got offered a job for the position of an AI developer and since then, I have been working in the same domain.
How could AI help combat climate change?
I think we can use the object detection and localization power of deep learning systems a lot to combat climate change.
With the huge amount of image data available to train deep learning models, systems can be developed which can detect plastic waste with reliable accuracy, and combining it with drone technology, we can automate cleaning systems.
Furthermore, plastic waste poses a higher threat to our marine life, as we see steep incline cases of plastic found in the bodies of dead fishes in the ocean.
This problem can also be solved using deep learning technology and drones.
The surge of cloud technology and IoT has made it easier for us to deploy our deep learning solutions on cloud services and have drones connected to them even while they are deep in the ocean.
I think this is one of the most amazing applications of AI, and I would like to see many more applications of AI in this field.
Deforestation is a major threat we are currently facing as a species and the testimonies of fires in the Amazon rainforest and Australian wildfires show how grave the problem is.
I think this is where AI is finding one more application, that drones can be used to plant trees by giving them the digital intelligence to traverse a region and use AI to find the most suitable geographical landscape to plant a tree and then instructing the drone to plant the seed.
Many surveys are emerging that this is cheaper than planting trees by hand and it would be nice to have such automated systems helping humanity and planet earth.
What goal you would like to reach with AI and Machine Learning?
AI as a field has emerged as revolutionary technology because of the advancement of the processing powers of computers.
We are developing better algorithms every day in the field of machine learning, deep learning and reinforcement learning to the extent that we have developed AI that can train itself.
But to date, we have not been successful in developing the values of trust and compassion between AI and humans.
I would like to see research done in that domain, as I think it’s vital that AI prioritizes the importance of human lives over its own and not be a threat to humans.
I would also like to see the applications of AI in healthcare, particularly in treating diseases like cancer.
There are projects like CRISPR which research in gene editing and they can leverage the power of AI to detect cancer cells and treat them.
At the same time, I would also like AI to be used in interplanetary travel flights and to discover planets that can sustain human life and civilization.
Do you have any AI projects that you want to share your experience?
Previously, I have worked in the domains of computer vision and natural language processing and I am very interested in various algorithms such as convolutional neural networks and sequence models.
Currently, I am working on recommendation engines, a domain that I haven’t worked on earlier.
I find recommendation engines quite difficult to work on, but at the same time, I am very enthusiastic about it as the quality of recommendations is subjective and we have to use techniques like AB testing to examine how well these engines are in the real world.
Furthermore, there are many tradeoffs in applying any recommendation engine algorithm such as the amount of user purchase history or the number of products to fetch recommendations from.
Besides, I am very much interested in learning GANs (Generative Adversarial Networks), networks that can create new data instances that can resemble your training data.
GANs are an exciting recent innovation in the field of machine learning and I am truly amazed by the applications GANs are used by the researchers, especially Text-to-Image Synthesis, image super-resolution and face in-painting.
These applications of GANs can be used a lot in surveillance camera footage editing.
Surveillance camera footage can be used as evidence against a suspect but sometimes, there are cases that the faces of the suspects cannot be identified as they are too blurry.
I am interested in developing a system that can at first detect the faces in a particular frame in the video feed and then use a GAN to increase the resolution of that face.
What kind of industry are you excited to see Machine Learning implemented in?
I think machine learning is a revolutionary idea whose power can be leveraged to perform quite a few extraordinary tasks.
I am excited to see machine learning algorithms used in domains such as software development, art, and education systems.
Nowadays, research is going on to develop machines that can code on their own without human assistance.
Microsoft’s Sketch2Code is an AI system that can generate HTML code of a webpage given a hand-drawn design.
I think this is very vital research, as these algorithms can help developers in software development by a large margin.
Software development is an iterative process and many times, there are changes to be made in developing the code, maybe due to the changes in requirements given by the client, conflicts induced in software libraries or some other reason.
This increases the time and costs on both sides. I think machine learning can be employed over here, that if we can use AI to develop and revamp code according to the changes in the requirements of the clients, we can essentially develop a very cost-effective software development solution.
Furthermore, I am interested in ML being used to write and proof-read textbooks. Recently, there has been research done where ML has been to write books and it is seen that these algorithms are very capable of writing sentences and linking multiple sentences to form a context.
Hence, I think we can deploy ML to write textbooks relating to a subject as well as proof-reading them in some way by checking the credibility of the sources mentioned or checking the facts and figures.
These are some of the domains I am excited to see more and more machine learning algorithms applied in.