Born and raised in San Diego, California, found my true passion for data when I studied Economics, Political Science & Math.
My fields of interest include : NLP, data visualization, Machine Learning , High Performance Computing,
Cryptography, blockchain, decentralized currency & many other fields including Math, Physics, Political Economy, Psychology & Philosophy .
A still more glorious dawn awaits / not a sunrise,
but a galaxy-rise / a morning filled with 400 billion suns /
the rising of the milky way.
- Carl Sagan
In the future, traversing galaxies and colonizing new planets will be all but an afterthought. The future of space-travel, will not be possible if we cannot predict
the trajectories of all objects in space, & their interactions with each other .
The following is a high-level presentation I did after researching the Fast Multipole Method algorithm.
As my interest in Machine Learning piqued, the problem of computational
complexity became increasingly apparent given the amount of data needed to run ML/DL models effectively.
The problem is best defined as the n-body problem: In physics, the n-body problem is the problem of
predicting the individual motions of a group of celestial objects interacting with each other gravitationally.
The n-body problem was first devised by Newton when calculating the trajectories of Jupiter & it's moons.
By 1987 Greengard & Rokhlin published: A Fast Algorithm for Particle Simulations*, proposing a new method of
reducing computational complexity from
By the year 2000 their algorithm was named top 10 algorithms of the 20thCentury.
What is the N-body problem?
Before getting into the nitty gritty, it would be a good idea to define the problem and why it is important...
Problem: Predict the individual motion of celestial objects & their interactions with each other.
A very simple way to think about this is to imagine someone juggling 2-3 objects. The task is simple for 2, but for 3
one needs to master hand-eye coordination & understand the mass & velocity of the object being juggled as well as the
force needed to move them.
The objects are easy to track given there are only 3
But Space is Big...
If we count the amount of stars, we are taking into consideration about 1021 stars, 1 Billion Trillion,
if your math is fuzzy. This is only the amount of stars, lest you forget there are planets, asteroids or the little understood Black holes.
The number is unfathomable, calculating mass & predicting the movement of each body & its interaction with other bodies is computationally impractical.
My test page
Lost in space are you? Yeah, same...
Solution? Maybe...
Enter Barnes-Hut Algorithm
Stated simply, what the Barnes-Hut algorithm does is create a quadtree (or octree) that computes close-range particles mass & force individually,
while distant particles are approximated by the largest macroparticle's center mass as well as force. This in turn dramatically reduces computation due
to calculations taking place as an approximation of particles instead of direct summation of all particles.
Quadtree bounding occurs where there are particles in a given space, otherwise leaves the rest uncalculated.
FMM
Many similarities to BH, but...
In essence, FMM & BH do the same computations, however, there are some differences. The first difference between
them is that FMM computes the potential at every particle, not just the force. The inner expansion also uses the nearest-neighbor to calculate potential.
The key difference is that FMM is more expensive because it computes both expansions (inner & outer) for every leafnode and BH does not.
Barnes-Hut vs. Fast Multipole Method
FMM wins at 10 million iterations Another Win
Conclusion: Thoughts & importance
Now that we have explored the significance of the n-body problem & some its proposed solutions; the question is : Why is this important?
A tidal wave of data is coming, 5g, driverless cars, smart homes & any internet connected device will put a strain on current & future hardware.
The necessity for faster & more efficient algorithms is needed for exascale & some day zettascale computation, the importance of these algorithms are a matter of national security. I modestly propose the idea of "computational supremacy" : The need for nation states to own & control powerful supercomputers
which will ultimately be used for military purposes like obtaining Strong Ai & mapping the cosmos.
The future belongs to the next generation of thinkers
Check out the TL;DR version w/ citations and applications of FMM
Online Appearances
Check out some interviews I have done for Google-Kaggle
Kaggle Days Paris Interview
Kaggle CareerCon
Data Science: Intro
The Titanic: An intro to Datascience
The 21st centuries sexiest job has been awarded to
data science
, the title is over the top but its importance is not.
About 4 years ago, I decided to change my entire life's
trajectory & learned how to code in Python, specifically for Machine Learning.
If you're curious, and willing to experiment,
understanding how to write code is one of the most rewarding experiences I have come to cherish. I wrote this short & simple
introductory notebook for people interested in data visualization and building simple machine learning pipelines for people with non-technical backgrounds..
Kagglenoobs
A Data Science Community
Kaggle
is known as the 'Home for Data Science' where the worlds most difficult Machine Learning problems are solved
for prizes. In 2016, I started a slack-channel for new Data Scientists who use Kaggle, I named it Kagglenoobs.
Today we are at ~11k members who range in ability, from the worlds best ML/DL Engineers to newbies who want to get
started in data science. The community is for anyone interested in Artificial Intelligence, paper-discussions &
interviews as well as coding help. The most rewarding aspect of my “accidental success” is that I am able to befriend such
talented and passionate humans. The Kaggle community is very welcoming to all walks of life, and most importantly
a place for learning.
The following are excerpts from over a dozen interviews I have done:
Q: When did you first come across Data Science and how did you know this was the way for you?
Anthony: My first job out of college was forecasting GDP, inflation and unemployment. I loved playing with data!
Every dataset has its secrets and I think it’s thrilling to try and uncover those secrets.
Q: What are the must know skills if you want to succeed in Kaggle? ie Blending, stacking etc. Also, as a student I
generally don’t participate in challenges that have huge datasets the main problem remains of the hardware. Would
you be open to providing free hours on Google Cloud?
Anthony: I suggest reading the winner’s interviews on the blog. You’ll learn from far smarter people than me! We hope to be able
to make more compute available to our community now that we’re part of Google, particularly for the bigger competitions. Still a
work in progress.
Author of Keras & Staff Software Engineer at Google
Q: Can you tell us about yourself, where you are from, where you live now, where you work, and what your educational background is like?
François: I'm a software engineer at Google in Mountain View, CA, USA. I mostly work on Keras and TensorFlow, and I do so
AI research as well (I used to do more, and I want to do more again in the future). I studied engineering and computer
science in Paris. I also spent some time in Tokyo where I was doing cognitive developmental robotics
Q: As a product manager for developer facing platforms (Kaggle and now Stack Overflow), you've frequently been an inspiration
to me! What is something you wish every PM for developer tools knew?
François: The key thing is to keep the user in mind: to ask, what's the value-add of a given new feature for users, how users will
discover it, how they will learn to use it, whether it will fit well in the workflows they're dealing with, whether it fits well with
the existing API and model models underlying the product, etc. Too often technical features are added with too much focus on technical
specification ("it needs to do this and this") and little regard for the experience of those who will actually use it
Q: Kaggle competitions for example seem apprehensive and I've been delaying taking part in this for over a year.
So what do you suggest one should do in this case? How does one get ideas? How should does one make the leap from studying
to doing a project?
François: I think Kaggle is a great place to start. It gives you a goal to achieve, it gives you data to work from
(so you don't have to handle data collection and annotation yourself), and you get access to working code examples if you're
not sure what techniques you should use. So doing Kaggle at first is a great strategy. After you get familiar with the process
of ML, just work on something you're passionate about that also has value for the rest of the world. Iterate quickly at first
(prototyping something and then moving on to the next project) until you reach something that feels like a breakthrough
Q: What do you do at Airbnb? How is it different from PetroBras?
Gilberto: "At Airbnb I worked for the search ranking team and right now I’m working on some exciting projects related to
Paid Growth (Ads basically). Airbnb is a tech company and Petrobras is an oil/energy company. So they are fundamentally different.
But the main difference between both I can resume in one word: speed. At Airbnb things goes very fast in all aspects."
IBM - Distinguished Engineer, Machine Learning, Optimization
Q: If Deep Learning is not the Answer to Ai/ML, what do think you think it will be?
Jean-François:
Tough question!
DL is probably the answer for perception automation. But this is not AI. People are ecstatic
because a NN can do what any 5 year old kid does without thinking. AI is broader, and I don’t
think we’ve made progress in general since 20 years. This is a long way to say I don’t know at all.
We are lacking the physics of cognition, i.e. a set of mathematical laws that we can use to design
new artifacts. There was some recent work linking NN learning with information compression. That’ s
the kind of progress we need."
Q: A common question that I ask in every AMA (we take this as a feedback actually): What do you think
of KaggleNoobs? What are some good/bad things about it and how can we improve?
Jean-François: I love it! I learn a lot actually, this community is fascinating. Lots of different background, cultures, skillsets,
interests. For instance, I am not a manga or anime person at all, but there are true experts here!
😂
Same for
video games...
NLP
something here
Submarine
Lorem ipsum dolor sit amet, consectetur adipisicing elit. Mollitia neque assumenda ipsam nihil, molestias magnam, recusandae quos quis inventore quisquam velit asperiores, vitae? Reprehenderit soluta, eos quod consequuntur itaque. Nam.
