RSSAmplifier

Blog

Iota

rohitghosh.github.ioRSS feed ↗10 posts

Latest posts

So Did You Come Up With A Novel Idea?

So Did You Come Up With A Novel Idea? I think I have come across this wordplay quite a bit of late: this is a “novel idea” or “can AI create a novel thesis”? Two things come to mind when I hear this. First is the obsession we have with novel ideas, especially among founders and researchers. Second is: what is even a novel idea, how do you define it? In 3, 2, 1… I dive down the second rabbit hole.…

To Preserve Or Not To Preserve

To Preserve Or Not To Preserve Of late, my wife has started to put these nice, handwritten notes on my lunch boxes that I really enjoy and adore. But a new challenge has come up with that: the dilemma of whether to throw them away or preserve them for eternity. Not sure about eternity, but long enough maybe? At least photos? Growing Up Without Keepsakes Growing up, I have never been much of a…

Are All Sports (Watching) Created Equal ?

Are All Sports (Watching) Created Equal? People have different likings for different sports—a lot of it depends on upbringing and surroundings. If you live in the desert, chances are your liking for ice hockey might be lower than for field hockey. But it’s worth pondering: beyond your favourites, are we predisposed to liking certain sports more than others just because of the sport itself? This is…

Visualizing and debugging deep convolutional networks

Have you been ever stuck while the neural network executes smoothly across epochs but the loss doesn’t decrease ? Or may be the loss decreases but not the way you would like it to ? Debugging a deep neural network is helpful if not essential for a lot of use cases. If you want to know why building a visualisation algo is important, especially in radiology - you can have a read here In this series…

Segmenting Overlapping Chromosomes using Deep Learning

After my last post on optimization of earnings by cab drivers as they drive through New York City – this post explores using deep learning aka Artificial Intelligence to segment out overlapping chromosomes on slides used for cytogenetic studies. The project is part of a set of open problems posted on AI ON website . The repo for the project is also available on my GitHub . We’re still working on…

Optimising Cab Routes In NYC Using Data

These days, on my cab rides, I have met some interesting folks driving the cabs. And, a lot of those times, when the driver was owner of the cab, our conversation tended towards to the ever-so-important question - “How to generate maximum revenue?” Do I drive at night? Where do I start my trip? What do I optimize for – longer rides in a longer time or repeated shorter rides in the same time? How…

Is unsupervised learning the way forward ?

Of late, we have heard a lot of talk about unsupervised learning being the way forward. Before we delve further, let’s clear the air around the types of learning. Supervised learning : Learning, wherein the model is provided with target labels. The model is trained to predict the labels. E.g. Detecting digit from a hand-written image, given similar training data with labels (i.e. mapping of every…

AI - Beyond Buzzwords & Comparisons To Human Abilities

Any heated argument on AI these days soon ends up in a predictable trajectory - so DeepMind’s AlphaGo beats Go champ, Tesla to compete with Uber using AI driven cars, ethical dilemma of driver-less cars and finally - would robots overpower humankind (on sidelines of the last debate would be venue of the Matrix themed D-Day - Earth or Mars !) So if you’re writing a sci-fi book or you’re a…

Why Tinder works? Solving for X in Tinder for X (Part 2/2)

In Part 1 , we delved into the user psychology of what makes Tinder such an addictive product and what are its takeaways for your business idea in general. We’ve all heard a lot about the Uber for X model now, where X has been abused and overused a lot (2016 is an especially interesting time for these startups). Let’s talk about the other cool thing - Tinder for X. So we ventured out to find what…

So, can coding be a lot easier ?

So most of us have had our share of interactions with developers, in some way or the other. Luckily enough I have now been on both sides - and frankly the experience of creating things from scratch as a developer has been overwhelming. But then not being a ‘techie’ from Day One has left me pondering with certain things, and in this “Hello World” of LinkedIn Posts, I take up one of them. The best…