"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

April 14, 2022

Federated Learning - How - Why - When

Summary from Quick 5 mins Tutorial

  • Client trains on data available at the device
  • Decentralized data 
  • Start with a model shared from server to clients
  • Clients which have sufficient data / Models deployed to them
  • Trained on local data and model sent to the server
  • Weights / Biases are shared with server
  • The server averages all the weights and creates final model
  • A collaborative and decentralized approach

Link - Session

Questions / Next Steps

  • Server Configuration, Tools, Package required
  • Client Configuration, Tools, Package required
  • How to train / run 
  • How the model gets updated between multiple clients
  • Similar to data synchronization need to investigate on infra needs to run

TensorFlow Federated (TFF) is an open-source framework for experimenting with machine learning and other computations on decentralized data. TFF runtimes to become available for the major device platforms

Code Example - Link

Tensorflow Federated Tutorials

Code - Link

Observations for code Walkthrough

  • Federated learning requires a federated data set
  • TFF repository with a few datasets, including a federated version of MNIST
  • Would simply sample a random subset of the clients to be involved in each round of training
  • Constructing an instance of tff.learning.Model

Research paper - Communication-Efficient Learning of Deep Networks from Decentralized Data

Key Notes

  • Decentralized approach Federated Learning.

Ideal problems for federated learning have the following properties: 

  • Training on real-world data from mobile devices provides a distinct advantage over training on proxy data
  • This data is privacy sensitive or large in size (compared to the size of the model)

Federated optimization has several key properties

  • Massively distributed, Limited communication 

There are two primary ways we can add computation: 

1) increased parallelism, where we use more clients working independently between each communication round; and, 

2) increased computation on each client, where rather than performing a simple computation like a gradient calculation, each client performs a more complex calculation between each communication round

Tensor Processing Units (TPUs) are Google's custom-developed application-specific integrated circuits (ASICs) used to accelerate machine learning workloads

From Link

  • Differential Privacy - Adding Noise to Ensure Privacy
  • Secure Aggregation - The server can only see bulk updates
  • Privacy is paramount in Federated learning
  • IID - Independently Identifiable Data
  • Privacy and Fairness are in the opposite direction

CPU vs GPU vs TPU

  • CPU - Small models with small, effective batch sizes
  • GPU - Models with a significant number of custom TensorFlow/PyTorch/JAX operations that must run at least partially on CPUs. Medium-to-large models with larger effective batch sizes
  • TPU - Models that train for weeks or months. Large models with large effective batch sizes

Federated Learning in Vision Tasks | Umberto Michieli, PhD@Uni of Padova, Intern@Samsung Research

My Feedback - let's collect minimal data and build models, Before we start to run, let's learn to crawl and walk :)

Keep Exploring!!!

April 12, 2022

How do you succeed in building impactful Data Science Use cases / Solutions ? Beyond Kaggle things to Learn ?

How do I find the most impactful use cases and have quick wins? Some guidelines / potential questions to give you the perspective.

I have participated in Kaggle and achieved a good ranking. I have a good understanding of Data Science, Let's build solutions. If all you have is a hammer, everything looks like a nail. Let's see beyond Kaggle what things we need to understand.

Impact #1 - What are the current challenges / business problems ? Identifying impactful / Potential Ideas ? 

Solution - Collaborate work with your business to understand, and get their vision, and priorities. Your use case has to be aligned with business needs / current challenges they are solving. A measurable ROI will always help to prioritize and deploy it to production.

ML Applicability #2 - Is this a Data Science use case, Does this need to change/introduce a new process, introduce new touchpoints, or is it a data or data science problem or Insights

Solution - Apply your domain lens, Data science lens, and take a transparent decision. Don't over-engineer for sake of it. If it makes sense do it.

Data Availability and Readiness #3 - If the first two parts are true, you spot problems, you see the feasibility of data science, evaluate what minimum you can build with the available data

Solution - Work with your Data/BI team, partner to build the required data for your MVP solution. The gap between reality vs expectations, What more data do you need to add more, integrate into the system you will get the clarity in this step.

You need to potentially collaborate with the business, product, and data team effectively to spot a successful opportunity. A lot of collaboration, and teamwork to spot the best use cases. Apply these questions and spot your opportunities.

Kaggle and other learning platforms work on the aspect of Feature Engineering, Model building, Beyond Kaggle this is the reality you need to look to apply Data Science in practice. Data science is #Teamwork. You need multiple lenses and participants to work to build impactful use cases.

Feel free to add other questions/guidelines as well.

Keep Exploring!!


April 02, 2022

My Perspectives of IT and Business Team Collaboration

If all my ideas have worked I would have climbed a few steps more. For the things it worked, Why it works? Why does it take so long to sell an idea? Does collaboration really work in action vs wishes of leadership expectations? To understand I wanted to look back at some paths in my history.

Sometimes when we look back we know why certain initiatives took so long, Why it works the way it works. This goes back to my Microsoft days XBOX supply chain Team. I was working with Program Management, Product Management, Support, and QA Team. The Product Management Team works closely with the Business Team.

Every Team has its own priorities.

  • Product management - Support Release Xbox, Xbox360 new type consoles with required code changes
  • Business - Setup new plants, identify new vendors, reduce Red Light Repair issues, Increase Warranty
  • Support Team - Reduce the number of customer issues
  • IT Team - Support all the priorities for business / qa / features etc

One issue of the customer writing an email about the warranty in the system is not correct. After a few executive-level escalations. It boils down, We spot there are issues with the way we store. To sort out the issue we provide a free warranty.

When I moved to Amazon, the Initial few days I was listening to another Team debating a similar workflow of repair order flow. I was thinking that everywhere edge cases are the big discussion items.

As we go along keep fixing, and adding new features. This system was one good enough system that tracks cradle to death of Xbox. Manufacturing, shipment, repair, fulfillment, sales, warranty, refurbishment, scrap. Everything was there about the console.

Every time when I hear these in JDs / Ecom Supply chain / Repair functional roles. I feel long back we had done all this as an in-house product. 

The core of it is warranty tied to a customer is a much better approach than a warranty mapped to a console. The order of transactions when they are out of sequence is created out of sequence status and we could keep track of update sequence vs delayed delivery of transactions.

I was able to work with my mentor Roji to build a prototype. It took so long release over release to recommend it to put in production. It had data migration, and a core engine but overall the changes were worth it. After a year when there were no priority projects, An initiative like that picked up this item. At least I didn't see any other announcement of a free warranty extension :) post-implementation.

I was wondering why did it take so long, Idea is necessary, we know dirty records in the system due to the sequence of transactions. There is effort and impact. The priorities across each function, business vs selling / pushing new ideas across functions take time.

What other options could have made it work?

  • If each function thinks about its own priorities it is not effective collaboration.
  • Sometimes when we solve a problem we need to think beyond fixing the current fire.
  • Everything is driven by a cost center. Fix now, Move on let's check later. The perspective of how the next customer should not go through the same issue needs a collective lens, not a function-based
  • Innovation in a way is looking beyond daily functions and seeing what matters more from a customer perspective
  • Keeping every function busy, planning back to back new items vs slow and steady everything has its own tradeoff/results
  • Sometimes even pushing idea visibility with title / connecting with leadership helps
  • If every team thinks from a customer point of view there would be more consumer-focused collaborative efforts than thinking based on individual function priorities.

The true essence of collaboration needs closer coordination, understanding, and planning across functions with customers as the epicenter of focus. 


Keep Thinking!!!


March 30, 2022

Storytelling Tips

Tips

  • SOS - Sense, Observe, Store
  • RSI - Real, Specific, Incidents
  • Stories of Self
  • Stories of Outsiders

Perspectives

  • Connect with person
  • Then build your story
  • Build your brand / Narration
  • The hero of the story is the audience
  • The story connects more with data

Personality

  • Self-awareness / Brand Building
  • Self-awareness
  • Key takeaway - Two styles 

Aristotle Principles

  • Ethos - Credibility
  • Pathos - Relate
  • Logos - Data

Keep Learning!!!

March 25, 2022

Cracking the code of 10 minutes delivery, Decoding - Data, Intelligence, Science behind it and Impact

If I get my favorite dish in 10 mins, It would be an awesome moment, the gap between wish vs reality is getting too narrow. Let's analyze closely the data, pattern, and science that bring reality to the promise of 10 minutes delivery.

This timed delivery has made a dent in history too. Domino's had great growth with 30 mins pizza delivery promise when it was competing against pizza hut in the late 70s (1979). This idea of the timed, same day, 30 mins had evolved across Amazon Prime, same-day delivery, swiggy instamart.

This worked best / works for groceries / retailing. 

How does this work? we will look at the pieces of the puzzle that eventually connects the solution.

As a retailer, for example, Amazon, I would know

  • Top selling items in each category
  • Forecast for those items by region / city / etc
  • With forecast/demand aspects I would keep a reasonable amount of stock preordered with essentials / all-time items / seasonal items
  • Most of the warehouses would be available near metros / little away from metros
  • In order to fulfill same-day delivery, you will have to keep some stocks based on future estimates
  • This same-day delivery may be available more for products of certain price range / Higher probability of sale / Vendor warehouses are available within the vicinity
  • Assume the stock is available, the next big thing is delivery

Routing

  • Logistics, Delivery, Routing/packing/delivery is the key
  • With Options to plan based on traffic patterns / leveraging on-demand vendors/availability of labor this delivery is also feasible
  • There could be multiple delivery schedules in certain routes to accommodate this dynamic same day delivery
  • The delivery I have observed is done with two-wheelers, 4 wheelers to cover both times, handle traffic, complete more deliveries

Instamart

  • Many times when we walk near HSR max, you will spot a three-floor building without any details and a lot of swiggy delivery boys waiting and collecting items
  • They are spaces that act as fulfillment centers within the city to deliver items within 30 minutes

How 10 minutes delivery is possible in Food?

  • Compared to other options Food has other factors #freshness #quality
  • Based on previous orders/customer patterns, Apps providers would know peak demand area / repeated buying customers, frequently ordered items
  • A few places where I have seen freshness handled is STAR Biryani. They prepare Biryani at 10AM, 1PM, 5PM, 7PM, They prepare food in multiple time slots based on demand as well as change based on observed behavior
  • This does results in repetitive work, labor, working round the clock to serve #fresh, and meeting #quality
  • Today we spot a lot of cloud kitchens, This 10 minutes delivery can very well be handled within 2 to 3 km of the vicinity
  • Forecasting, Allocating delivery specifically for these orders, Shortest routing approach based on current traffic, Most of this is already handled in their current App for the usual 30 mins normal order delivery
  • Somewhere during the presentation, I captured this screenshot, Zomato / Swiggy know both consumers, restaurants very well.


Caution - Food Quality

  • My personal observation, Out of 10 places only 1 or 2 places I have been a repeat customer, What we see in Pic vs What gets delivered vs Health issues is always a warning
  • During my childhood monthly once/twice was our usual restaurant dining routine, it was expensive and not affordable
  • The 20s/30s We would be able to explore/experiment with food but the preservatives, artificial color, a sedentary lifestyle will reflect the consequences in the 40s
  • This ready-made food consumption in the long term when it exceeds the limits will be a health hazard, Garbage in Damage Done :)
  • The target segment would be primarily working segment / above middle class, Easily today for 2-3ppl you would spend 1000+

Everything, when it exceeds the permissible limit, will become an addiction in long term. 

It was a little long story, let's get back. 

I was thinking of another business idea. If I could prepare an API-based approach where we could directly provide the recommendations/promotions to vendors we can make it more R2C, Restaurant to consumers directly than R2A2C Restaurants to App Providers to Consumers.

I hope some perspectives were interesting :)

Keep Exploring :)

March 20, 2022

Metaverse and Computer vision

I was listening to Lex Fridman and Mark Zuckerberg's podcast on metaverse. Some insights/ideas I got out of the podcast. In a metaverse environment, the goal is to make it closely mimic your physical presence by paying close attention to representing/capturing your real self. The key vision use cases that would play a key role in a real-time environment/experience for users.

  • Emotions tracking
  • Facial Expressions
  • Tiny gestures/remarks unique to the personality
  • Face tracking
  • More realistic presence for touch/feel senses
  • Your AR / VR device is going to be enhanced to track these details

From a computer vision point of view

  • Creating a deep fake representation/photo realistic / cartoon representation of yourself
  • Representing your emotions, lip movements, expressions
  • Adding your voice modulations in discussions
  • Altering your face to sync with real-time expressions and movements
  • Background changes, clothing, and presentations
  • Creating your avatar for business, entertainment, family different groups

Revenue approach?

You might end up using it for different use cases like

  • Gaming / Entertainment / Business / Collaboration Multiple avenues of opportunities
  • Buying / selling / clothing / styles / accessories in Retail / Fashion domain
  • Influencer-based sales/advertisements.

This could potentially lead to a long-term subscription-based model.

Today with affordable smartphones, high-speed connectivity Spotify, youtube we spend lots more time than 10 years back. In another 5 years, These could be replaced with affordable 50$ AR / VR where we might split our time spending across social media/youtube/metaverse, etc..

Keep Exploring!!!

March 18, 2022

Gartner Hype Cycle Observations..

 

  • Plateau of Productivity - Predictivity Analytics
  • Enlightenment Stage nearing Productivity - Text Analytics, DBMS Analytics, RDBMS with Analytics APIs
  • Currently Setting / Approaching Standardization - Video Analytics, MLops
  • Innovation Trigger - Feature Stores / Synthetic Data / Reinforcement Learning

Invest in Talent, Learning, projects based on business need, impact, maturity of the architecture

Keep Exploring!!!

March 14, 2022

AI Noise Cancellation

I hadn't tried it successfully, This seems to be a far better solution in meetings. Finally caught up with it. I had an Nvidia GPU Machine.

Installed Nvidia RTX Voice. The system device input remains the same


Added Nvidia RTX Voice to use the default input and cancel the incoming noises


Since RTX voice will get a stream of input and cleared voices, the same can be configured for other apps like Teams, Chrome etc..This is teams settings


This is chrome setting


Time for noise free discussions :) 

Keep Exploring!!!!


Experience vs Algorithms vs Design vs Polyglot expertise

  1. If I solve all algos and core data structures - Does it make it a good programmer - Yea Possibly he can solve build solutions
  2. What do I do in my work ? - Understanding data, domain, customer problems, applying the lens of data + ML + BI finding potential solutions 
  3. Where does this experience come from? - Similar domains, problems, building products 
  4. What does experience mean ? - Collection of different roles / functions / projects / products 
  5. Did I do only development or support or testing or performance? - When you build solutions you have to wear multiple hats to build them. There is no hard boundary for each role. To think from a customer perspective and building solutions is different from building solutions and how customers will use it
  6. Do I remember all algos, code now ? - Now, Some I remember, Some I learn as I apply
  7. Do I need to learn to practice every day? - There is no boolean way of answering for knowledge to say you know or don't know. As long as you can build solutions and code up you are good enough to solve customer solutions.

There is no one definition of skills. Do not go by what is being dictated. Building solutions takes as much time as you learn core cs basics. Expertise comes with time and experiments, not just coding standard problems.

Keep Thinking!!!

March 06, 2022

Boolean System of Education / Learning

Always we get evaluated for success/failure. There is no place for passion/experimentation/work on something in long term. 

  • Pass / Fail
  • Engineer / Doctor
  • Data Scientist / DB Developer

The flaw that you get graded and evaluated fit or fail is itself wrong. How many times have we passed an exam but had to relook to learn a concept? From applying/solving Data Science use cases vs when I had to revisit basics. I had to relearn everything because In real-world scenario data/use case / ML Algo / Accuracy / Deployment those take-up time and focus, not first principles. Balancing both building blocks vs implementation needs time. 20 years gave me a perspective of how it evolves. 2 years of master's helped me unlearn/relearn. Still, I try to bridge the gaps. Sometimes the questions help me to find another convincing answer not a namesake answer. Multiple choice questions can deem you as failed but your perspective may be beyond those MCQs, When you learn a skill some things could be intuitive not remembered from an exam point of view. When you do several hands-on experiments, you can relate/connect better? What you read/see/hear may not connect well.

  • Learning is endless and it cannot be evaluated with point-in-time marks/evaluation.
  • Good communication skills do not mean good technical skills
  • Certifications do not mean you have the required skills
  • Titles do not need to reflect competency 
Everything is a combination of skill/opportunities/timing/exposure. Everyone has their own time to learn/grow. You just need to be sure Am I better than my previous version. 

Life seems too short as I keep stepping into the 20+ exp zone. Follow your path, retain your uniqueness.

Experience = Collecting different experiences/roles/projects, Not working on the same things for X number of years. Collect more memories to connect/relate and build the big picture. 

Keep Thinking!!!