Federated learning & Security (Story of a CEO & 2 researchers)

Hello Followers & Data Friends,

Again, thanks to someone who wrote this in my book review and helped me to restart this blog again.. I am glad someone still remembers my old blog and so many got motivated to enter AI & Data Science after reading it. My humble thank you - 

It's Sunday night (11:30 PM) and amazing times that we live in where AI and Data is the new oil. And it's also the time to have a deep look at Sundar's last month's interview where he spoke about Federated Learning (starting at 27:50). (Yeah, it was another busy weekend with lot of work but I still listen to Sundar's speeches on my walk) - https://www.youtube.com/watch?v=EuF8nv53JeI&t=1997s

Federated Learning & Edge Computing is an evolving topic most popular since Google announced its cookie-less marketing world last year - https://www.gartner.com/en/marketing/insights/articles/three-steps-for-marketers-to-prepare-for-a-cookieless-world

Sundar also spoke about Privacy, AI and a lot of futuristic things.

The books/papers that came to my mind immediately after listening to the interview are -

1. Milind Tambe's book on Security & Game Theory (2011) - https://teamcore.seas.harvard.edu/publications/security-and-game-theory-algorithms-deployed-systems-lessons-learned

This is one game theory book I liked a lot in 2011 and wrote about it in my blog that time. It was fresh with good perspective security & game theory while it was written in a simple language. Also, amazing amazing chapter 12 on Stackelberg versus Nash in Security Games. Every game theory & security student should read this chapter and if possible attend Milind Tambe's lectures. His lectures have amazing clarity & awesome vision. (Yes. I don't promote my own books 😜)

2. Advances and Open Problems in Federated Learning (2019) - https://arxiv.org/abs/1912.04977 by Ramesh Raskar & the team. A wonderful wonderful paper on Federated Learning. 

The chapter I Like in this paper/booklet is - Adapting ML Workflows for Federated Learning, Hyperparameter Tuning and Neural Architecture Search (NAS) which is inspired by Chaoyang He, Murali Annavaram, and Salman Avestimehr' s paper on FEDNAS - Federated deep learning via neural architecture search. 

I will try to cover these things in my any upcoming guest lectures very soon. I have been extremely busy. For now, it's time to listen to Milind Tambe's all lectures from the past including AI for Social Good lecture including his JPMorgan lecture... 

More coming soon on this topic...

4 interesting papers on Ricci Flow & Deep Learning

Hey Folks,

Greetings. Its Saturday night and a new blog post from DataHulk.. Funny Funny name.. I came up with this name 12 years back .. when I get angry I work on data...! 

It's been some time since I spoke about Neural Networks, GANs & Ricci Flow (Ricci Flow was an important geometric tool in solving Poincare Conjecture....). Please feel free to check out latest paper on Shannon entropy power on Riemannian manifolds and Ricci flow - https://arxiv.org/pdf/2001.00410.pdf (Lot of hints to Information Geometry & Topology..) 

Here are 3 interesting papers which came out since my talks & articles (apart from Melanie Weber's papers at Princeton on Big Data & Ricci Flow.

1. Zhejiang University - THOUGHTS ON THE CONSISTENCY BETWEEN RICCI FLOW AND NEURAL NETWORK BEHAVIOR - https://arxiv.org/pdf/2111.08410.pdf

2. Oxford University - Over-squashing, Bottlenecks, and Graph Ricci curvature - https://towardsdatascience.com/over-squashing-bottlenecks-and-graph-ricci-curvature-c238b7169e16
3. University of Cambridge - RicciNets: Curvature-guided Pruning of High-performance Neural Networks Using Ricci Flow - https://arxiv.org/abs/2007.04216 

 

The computational graph is pruned based on a node mass probability function defined by local graph measures and weighted by hyperparameters produced by a reinforcement learning-based controller neural network. 

3 unique papers with 3 different ideas of using Ricci flow in deep learning. 

The 4th one which is very very unique is CONVERGENCE OF RICCI FLOW SOLUTIONS TO TAUB-NUT by Francesco de Giovanni - https://arxiv.org/pdf/2008.03969.pdf  

It does not show the applications through deep learning yet but as amazing paper which might have some future applications in deep learning.

Interesting case of Schrodinger equation & deep learning (and possible application in Commutative Hodge Conjecture)

If my readers remember my old blog (written under the pseudoname of DataHulk 😏), I used to talk about Physics Informed Neural Networks, Physics formed deep learning & Data, Machine learning & Deep learning to solve math conjectures 12 years back... much before The Ramanujan Machine (AI to Solve Math Conjectures) was born. 

And Yes, I saw review on my 2013 book - Healthcare Social Media Management and Analytics, so I know some of my readers have missed me 🙌🙌🙌





1. Here is a curious case which was written few months back - Data-driven vector soliton solutions of coupled nonlinear Schrödinger equation using a deep learning algorithm - https://www.researchgate.net/publication/355789067_Data-driven_vector_soliton_solutions_of_coupled_nonlinear_Schrodinger_equation_using_a_deep_learning_algorithm In this paper, there is pre-fixed multi-stage training algorithm by combining the error measurement & multi-stage training. This algorithm is much better suited for different dynamical behaviors of solitons with faster convergence rate.


2. Recently, G. Tabuada from MIT proposed a series of noncommutative counterparts all conjectures including Grothendieck standard conjecture, Voevodsky nilpotence conjecture, Tate conjecture, Weil conjecture etc. XUN LIN has also proposed NON-COMMUTATIVE HODGE CONJECTURE.


Similar physics informed neural nets or GANs can be used for commutative Hodge Conjecture. More to come soon on this topic.


On a different topic, Please do listen to András Juhász & Marc Lackenby (similar to University of Sydney mathematician Geordie Williamson's work with DeepMind on representation theory) - https://www.youtube.com/watch?v=hIUiPi-jAjM

Swarm Intelligence - Another curious case of Duck Swarm algorithm & New Google algorithm on Pathway Analysis

Hey guys, Thanks for reading my restarted blog. A lot of views and reads (that also on 31st night.. I must say I have very cool & eager readers.. I hope everyone had a great new year's night..  

Math is everywhere.. A lot of my readers might remember my old orchestration on Ant Colony Optimization, Swarm Intelligence & Quantum-behaved Particle Swarm Optimization algorithm (And a fun hint on How Ant Colony Optimization is used in CyberSecurity with reference to first Avengers movie dialogue - Nick Fury to Loci when he is in the prison- "the touch of the ant and the click" ...

And And And... funnier hint that Marvel might release a movie on Ant-Man etc..

A lot of advances have happened in Swarm Intelligence since then.... 

1. Quantum Particle swarm optimization (PSO), 

2. Firefly algorithm (FA), 

3. Chicken swarm optimization (CSO), 

4. Grey wolf optimizer (GWO), 

5. Sine cosine algorithm (SCA), 

6. Marine-predators algorithm (MPA)

7. Archimedes optimization algorithm (AOA)

8. And Surprise Surprise - Duck Swarm Algorithm 

Please check out 2021 paper - Duck swarm algorithm: a novel swarm intelligence algorithm by Zhang , Wen, Yang - South China University of Technology, Guangzhou 

A couple of simple images from the paper - 

And the Pathway Analysis based on duck swarm - 

By the way, A very very interesting thing happened a few months back. Google came up with their own AI tool on pathway analysis - https://blog.google/technology/ai/introducing-pathways-next-generation-ai-architecture/  

The question is would Google use more & more swarm intelligence in their pathway analysis?