Presenting to crowds, ML madness with tumour classifiers, and some brain food!
Vinaya Sharma | November Newsletter ππ¦
Hey! If youβre new here, hi Iβm Vinaya and welcome to Bits and Bytes! Iβm a 16 y/o emerging technology enthusiast hoping to change the world! π
In this monthly newsletter, I share bits and bytes of my learnings to help you grow from my key insights! π± Stay tuned for bits of self-improvement techniques, and bytes of groundbreaking tech innovations.
November has been all about performing, presenting, building and innovating. Check it out!π
Presenting to 150+ people
This month started off with an absolutely surreal experience at the TKS Explore hackathon. After just 4 hours of research, design, prototyping and pitching, my amazing team of Priyal, Tanvi, Zara, Krish, Shimoi and I built out Purify.AI. A new and improved 2-step municipal water filtration system, ensuring water is safe to drink for everyone. π§½π§
Currently, around 300,000 diarrhea-associated deaths occur among children in India each year. Not to mention the deaths associated with cholera, typhoid, polio and more. These are all waterborne diseases ruining lives due to the current ineffective water filtration systems. And this is where Purify.AI comes into the picture. Our 2-part solution includes:
Detection π΅οΈββοΈ. Through the use of microscopic cameras and computer vision (specifically convolutional neural networks), the hardware detects the exact bacteria present in the water.
Personalization π. According to the specific bacteria present, a current is sent out by electrode lattices installed in municipal pipelines to dissociate bacteria detected by the cameras. For example, with a 72 mA current, we can break down E-coli and successfully complete this process of collision-induced dissociation of ions.
We took home both the best prototype and best overall awards π, overall this experience was insane.
Now as I sat down to write my newsletter today I flooded the page with my key takeaways. As I represented my team on stage presenting to over 150 students, innovators, and judges, I had my highs and my lows, and I want to let you in on the journey. Some of my key takeaways:
Delegate π. You are working in teams for a reason. Assign different members with different skills, experience and interests to work on specific parts of the project. By distributing tasks you are able to go deep into a solution and not remain surface level with all of the components. For us, this meant breaking down the solution into 2 components, detection and dissociation. And breaking tasks further down into hardware, algorithms, diagrams/models and theory. By crowdsourcing knowledge, our team was able to hold each other accountable and get very specific.
Make it simple. Itβs pretty obvious that saying βwe are using computer vision to detect bacteria and electrode lattices to dissociate themβ, is hard to imagine. But perhaps a diagram and 3D model can explain the solution a little better. At the end of the day, the judges are assessing you on your ability to articulate your thoughts. If itβs complicated or far-fetched especially in competitions looking for solutions applicable in the near future, you must reassess. This is definitely what my team did best, we put in the time to design prototypes to tell the story and it helped to clear many questions! Β

But I wonβt lie to you, most of my growth happened moments before the pitch. As I sat at Torontoβs Design Exchange while the thoughts of my racing mind blew past me, moments before taking center stage. Check out this article to learn more about my pitching experience. π
Digging deep into CNNβs
I canβt imagine living without sight. Itβs honestly a superpower. With machines now able to see, I donβt even know what you can call them, other than magical. And so this month I set off on a journey, a hunt for a computer vision problem worth tackling. I wanted to create a model that impacted millions and so I decided on completing brain tumour classification models.
Currently, over 2/3 of people on earth do not have access to radiologists. 14 African countries have no radiologists, India has radiologists at a 1 to 100,000 ratio, and even developed countries includingΒ Canada and the UK have 30-day long wait times in order to get your medical images read. π = π
I programmed 2 models that detect whether a brain tumour is present and further classify the images into 3 of the most popular varieties. By programming both from scratch with Pytorch, and with transfer learning, I was able to apply my skills in developing AI pipelines and assess the benefits of the 2 methods.
Transfer learning obtained nearly 15% better results achieving a 90% accuracy on the testing dataset. π₯
With just a couple of lines of code, I developed a model on par with professionally trained radiologists. π€― This is mindblowing and really goes to show how Transfer learning proves to boast amazing accuracies when less data is available and when models exist for similar tasks as the one at hand.
If youβre interested in AI, Computer Vision, or building out some CNNβs, join me for a little Machine Learning Madness and come partake in The Epic Battle between Transfer Learning and Models from Scratch for Brain Tumour Classification! π§ π€Ί
Working with Walmart to improve the customer experience
Walmart is theΒ world's largest company by revenue, operating more than 10,500 stores and clubs in 24Β countries, under 48 differentΒ names. Walmart has built an empire with its everyday low prices, but it still has so much room for growth.
Alongside my challenge team at TKS, I am digging deep into how Walmart can improve the customer experience by developingΒ simple & scalableΒ in-store digital touchpoints or experiences that further connect customers to our digital journey. With the 21st century revolutionizing nearly every industry it is important that Walmart innovate to remain competitive. π
These past couple of weeks I have been researching, brainstorming and picking potholes into potential methods to revolutionize the retail experience in the near future. From 5:00 AM group meetings to external validation calls, this challenge has really questioned my time management skills. Unfortunately, I canβt share any of our ideas until next weekβs submission, but stay tuned for my next newsletters to break down our ideas to improve the shopping experience.
Some bits of self-improvement techniques π±
Letβs get to some mindsets! While building and learning itβs easy to get lost in thought, but remember the importance of intention. This month I trained my mind on self-growth techniques to optimize life. These are my key takeaways, and some of my favourite life hacks up to date.
Our opinions, judgments and impulses are up to us. We control our own thoughts and emotions. No one else can do that for us. When bad things happen, it's up to you to react in the most optimal way. Inner happiness is proportional to your values, so understand what those are and stand by them. π§
Teather your success to something you can control. Some people are just lucky, and some are very unlucky. Although you can make your own luck through hard work and determination, you can not control many external factors. Focus on what you can control and optimize for success in that area, stop wasting time complaining about things you can't control.πΉ
Stop being a slave to the opinions of others. With social media taking over more and more of our lives, we are being judged more. Social validation stops us from taking certain actions and causes us to embark on bad actions as well. Stop being a slave to others and do whatβs best for you! π±
We lose the most from inaction. Fear prevents us from reaching success. It is important to break down your fears and analyze what it is you are so scared off. Break down the fear, the worst-case scenario, and repairment techniques, this helps you realize your fears are not as bad as you think. With this wholesome picture in mind, you can decide if risks are worth taking, and you can tackle your fears. π»
Bytes of brain food π
Each month I consume loads of content through articles, podcasts, and videos. Here are my favourite stories that caught my eye.
On November 18th Nvidia released Magic3D, a text to 3D model generator. The output is a 3D mesh completed with textured colour. With modifications, these models can be used in video games, CGI art scenes and more. Magic3D was built in response to Googleβs DreamFusion announced in September, but this model is 2x faster. At the speed of these ML improvements, we will be seeing 3D content creation democratize and explode in the near future! πͺ
Did you know AI can detect Autism? 1 in 100 around the world is estimated to be on the Autism Spectrum, but what are the real numbers? Thereβs currently a massive gap in autism diagnosis as this neurological disorder is underdiagnosed. Currently, researchers are looking into Logistic Regression and Random Forests with the help of medical histories and real-world medical data to diagnose children between 18 and 30 months. Specifically, voice data identification is being tested out to help receive early diagnosis and potentially improve long-term outcomes for these children. π©ββοΈ
Stability AI has recently announced Stable Diffusion 2.0. A new and improved multipurpose version of the renowned Stable Diffusion. This release boasts an improved text-to-image model, image upscaling, inpainting tool and more. Using a brand new text encoder images can be generated to sizes of 512x512 and even 768x768. Your blurry images can even be upscaled by a factor of 4 now! π
Closing thoughts
I find it so exciting, almost exhilarating to wake up in the mornings now that Iβve found something I enjoy. Asides from the many code bugs, building in AI and seeing complete projects come together has been making my day, and I hope you to can find that piece of enjoyment that gets you out of bed every morning.
But asides from work, building relationships and having fun are equally as important. This month I performed at a Peel school engagement event, and a Remembrance day ceremony for more than 300 people, got to catch up with some old friends and of course, question my life while working with my favourites on 5:00 AM calls.
You killed the presentation π. Hyped for the next newsletter (can't wait to make another cameoπ)
You are incredible. Your progress this month was outstanding! Im so excited to see what December brings you :)