The Simulacrum
Published in

The Simulacrum

2017 : Aces Wild — 2018 : Blind Raise

At the beginning of 2017, I made a promise. To go big and audacious in goal setting as proclaimed here > 2017 - Big, Hairy and Audacious

I picked a goal each for mind, spirit and body. Here are my learnings and scorecard on how I fared. (Please leave a comment on how you guys did)

Mind — Learning

(Will try to keep this as non-technical as it needs to be) The goal was to develop a Artificial Intelligence Augmented Reality interface similar to Iron Man’s heads up display.

Starting 2015, Microsoft and Google had already beat human accuracy in image classification. This was in the iconic-image set category (where the pic has only 1 image to identify). The challenge I wanted to work on was in identifying all objects in a image from a real-world scene. 2017 was ripe with multiple projects in this area. When I started, I knew very little about vision computing.

4 quarters of preparation. It took an entire year of reading, coding, research and repeat cycles in vision computing and deep neural networks before I could even start to put together an interface. The cost of hardware and the APIs were restrictive for me to invest in a programmable headset. The neural net architecture that I envisioned that I would build, required a 8 GPU Titan-X module to even train a reasonable network and this was quite restrictive for me to move forward. Overall it would have costed me $20K if I decided to invest. I did not.

Instead, I procrastinated. I dabbled around with different vision computing approaches in small scale on a home Macbook Pro, training extremely small datasets (and failing miserably) in classification tasks for nearly 3 quarters. This failure was demotivating and disheartening to say the least.

I chanced upon a research paper titled “Focal loss for dense object detection” published around Aug 2017 that was ground breaking. A team of researchers from FAIR (Facebook Artificial Intelligence Research) had developed a novel approach in non-iconic object recognition based on Retina Nets. This was a turning point for my work. It took a lot of time to understand the pragmatic working of RetinaNets. I did make small-step progress and built my confidence back in the project.

Voila : I made progress by Nov 2017. Yup after 11 months, I could build and train a network that could identify object in a scene and can identify nearly 70 different categories of objects with good accuracy. This is no where close to what FB, Google, Amazon or Microsoft may have been working on, especially with their large investments. Think of my project as a first iron man’s suite that he built in a cave. I call this project Kena.

Here are few results:

Object identification and localization in a scene

Mind — Scorecard

Score: I would rate myself a 4 on 10 on this goal.

Where I failed: I am not happy. I could have done vastly better. I could have worked on a phone app (instead of waiting for the headset) to end the year with a app demo. I procrastinated the first-half of the year in analysis paralysis. Procrastination is an illness. I broke my own rule of “Action cures illness”. This is a hard lesson re-learnt. I promise to do better in 2018.

Where I did well: I am happy with the fact that I did not give up on the research though. I knew that I need to stand on the shoulders of the giants and not solve all problems myself. I did not break the rule of “Don’t be a hero”. This (though late) served it’s purpose of progressing on my project with a decent finish to the year. I am happy that I eventually succeeded in building a tight network that runs on a $2K laptop (1/10th the original cost budgeted) for the project.

2018 — What next? I will continue to invest time to develop a full blown application under the guise of project kena. The goal is to have a fully working application by end of 2018 that people will be able to use.

What do you guys think? Please leave a comment on how would you rate me on this project?

Spirit — Learning

This was a bummer. I proclaimed that i will learn guitar and play Metallica’s ‘Enter Sandman’ by the end of the year. Here is an account of how I progressed:

This is how I sounded on Jan 7th, a week into practice:

June 13th — 6 months in

Now, I am sitting here wondering if I should really upload how I progressed on Enter Sandman. It shall be blasphemy if I did. It’s Metallica for God freakin’ sake !! and my solos SUCK !! I will hold back on this until I am worthy, BUT, here is what I did instead.

I composed my own song called ‘Freedom of Puppets’ at the end of 11th month. I coded the drums on Logic, composed the riffs, lyrics and put together this draft version for you guys (I hope you have insurance). Listen to this :

Spirit — Scorecard

Score: I would rate me 6 on 10.

Where I can improve: This one is simple. More training. There are no shortcuts. Absolutely more diverse set of riffs and solos. I believe that I could have been less distracted with social life and put in more hours into training. I made some good progress for the first few months. I think I got carried away with my progress initially that I started slacking majorly after 6 months. The lesson here is to “stay the course”. I promise to do better.

What do you guys think about this song? How will you guys score this? Please leave a comment.

Body

Ok, I have been told not to upload pics ! :)

I started the year with finishing a 32 mile run in the hills. An account of that here > The Race

I needed to change my body type from a long distance endurance athlete to someone who can withstand high intensity weight and cardio. I changed my training schedule to add 85% high intensity weight training.

My deadlifts, bench and squats have improved to a good level. I started with a bench of about 90lbs and squat of about 140lbs. Happy to note that I can now bench about 180lbs and squat 230lbs free or about 550lbs on leg press. I have lost a good amount of fat and look lean.

Body — Score: On the looks, my kids rate the improvement 9 on 10, my near and dear ones rate the progress as 7 on 10. I would rate me 5 on 10. The reason I rate conservative here is that I have stopped running !! Yes, I nearly shifted from a long distance runner to just doing high intensity weights !! This is not cool. Also, I DON’T look like Tarzan at all :) I promise to change it in 2018.

2018 — Forecast

2017, I did OK. To be honest I am disappointed with the average show. This is not how it was supposed to end ! But then again, goals do not end.

2018, same goals, new vigor, quite some learnings from failure. I promise to do better. I can, I will…

You play the cards you are dealt with. Nothing more, nothing less.

Have an amazing 2018 and wishing you all a great success.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Freedom Preetham

Freedom Preetham

Founder Kena.ai, Cognit.ai | Math | Artificial Intelligence | Quantum Fields | Cognitive Science | Music | Poetry | linkedin.com/in/vvpreetham