"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" ;

July 21, 2023

Vision Interesting Startups

  • Virtual Try on Beauty- Perfect365, Orbo
  • Virtual Try on Jewelry - Tangiblee Virtual Try-On
  • Image Catalog Management - Vue.ai, Spyne.ai, Stylitics, wide eyes
  • In store analytics planning solution- Increff, JDA
  • Face Beautifier - Banuba
With Generative AI, Vector databases many of this will face tough competition to sustain



Keep Exploring!!!

July 20, 2023

3.5 years of AI / ML work and some key projects

Some key moments of the past 3.5 years of work 
  • Setting up / Driving a Virtual NRF - Driving everything from India
  • Strategy roadmaps based on the current state of multiple customers in  Retail, Logistics, Beauty, etc..
  • Vision-based products / Roadmaps / Production Architecture - Virtual Tryon
  • DAM / Stylitics / Try on Solutions 
  • Vision Products / Projects - Skin Care / Leaf-based plant classification
  • Forecasting projects - 300K products for Beauty product retailer
  • Bundle recommendations - For a clothing retailer, Figuring out what sells well
  • 2 months AI + Analysis + Troubleshooting a slow-performing Trading App (Air Crash Investigation type work :))
  • Image Search Engine, Image Catalog creation
  • A ton of tech reviews, architecture discussions
  • Code up as and when needed, Code / Learn / Handle both tech + business audience
  • Three batches of training AI / ML for 150+ product managers, Publishing POV
  • Reporting feedback/improvements for AzureOpenAI, Google partners, etc.
  • Currently in LLM, GenAI Mode
  • External Talk in One of the Conferences for Virtual Try on
  • Almost all good moments, except a few situations where I would be cautious if I spot such symptoms
  • Lot of coding/teaching for Upgrad :) to get better at basics :)
In the next 5 years how we code, will it be mostly prompt-based or idea based looks like a lot of new approaches in development :)
  • Tyco and Microsoft taught me a lot of domain
  • Even after a master's in ML, Domain knowledge + Common sense helps me more than ML views
  • Still could recollect key tables at least 100 tables in Reverse Logistics work - Product, Warranty, Msops, SST Tracking, Repair, Warranty etc..
  • The warranty rewrite work is still memorable and applicable till today
  • The 3PL touch point connectivity is still relevant in supply chain visibility
  • Sensormatic gave the RFID + EAS + People counting + Instore retail operations
  • The heart and soul of instore operations is based on store planning + planogram + store layout + real-time alerts + cycle counting + a lot of real-time opportunities with Vision
  • Now if I look back a lot of ML is applicable and I will rework If I have to redo those problems yet again :), Classifying the type of customer issues in XBOX, using NLP to address customer issues with GenAI, Past had a lot of data
  • Now all data can generate signals - Video, Audio, Text on top of RDBMS
Learned a few things (New start)
  • Optimization opportunities
  • Exploring pyomo / other relevant connected opportunities
  • Azure data curate features/masking/removing/compliance
  • A ton of training / long term - only to learn more 'persistence'
  • A ton of mentoring internship projects 
  • Reviewing/panelist in several events
  • Tyco was a bit relaxed with work, This is a marathon
Wishlist
  • Need a break!!!
  • Write a book
  • Take a break
  • Try building some ideas/products
  • Business + Tech is always essential to see the big picture / Explore options / Freelance
My perspective




Ref - Link

Why AI / ML ideas Fail in Large Enterprises?

Keep Thinking!!!

July 18, 2023

Streamlit Deployment Notes examples

runtime: python
runtime_config:
operating_system: “ubuntu18”
runtime_version: “3.8”
entrypoint: streamlit run --server.port=8080 --server.address=0.0.0.0 --server.enableCORS=false --server.enableWebsocketCompression=false --server.enableXsrfProtection=false --server.headless=true app.py
env: flex
network:
session_affinity: true

Keep Exploring!!!


Switchon - Vision Solutions

Vision for Automated Inspection 


 



Ref - Link


Keep Exploring!!!

Lightweight Deep Learning - Model Tuning - Model Compression

Key Techniques

  • Pruning
  • Weight Sharing
  • Quantization
  • Low-rank Approximation
  • Sparse Regularization
  • Distillation

Pruning Weights

  • Motivated by how real brain learns
  • Remove weights which ๐‘ค๐‘’๐‘–๐‘”โ„Ž๐‘ก < ๐‘กโ„Ž๐‘Ÿ๐‘’๐‘ โ„Ž๐‘œ๐‘™๐‘‘
  • Retrain after pruning weights
  • Learn effective connections by iterative pruning
  • Between L1 and L2, which regularization is better

Criteria for Pruning 

  • Minimum weight - Pruning by the magnitude of Kernel weight (L2 norm)
  • Smallest activation - Prune Kernels that lead to feature maps with the, smallest activations

Weight Sharing

  • Compress the neural network with weight sharing
  • Use a low-cost hash function to randomly group connection weights into hash buckets

Quantization

  • Binarize all the weights and activations, turning DNN into Binarized Neural Network (BNN), in order to reduce memory consumption and increase power-efficiency.
  • The method binarizes both weights and activations, in contrast to BinaryConnect which binarizes only weights

Low-rank Approximation

  • SVD

Sparse Regularization

  • Zero-out groups of weights using sparsity-inducing penalty

Distillation

  • Based on teacher-student model
  • Given trained deep neural network (DNN), called ‘teachers’, make a compressed ‘student’ model, with similar accuracy using quantization and distillation
  • Teacher = Original Deep Model, Student = Quantized Model

Ref  - Link


Keep Exploring!!!

July 16, 2023

Your weakness is also your strength

Appraisal for Horse - You run fast but you don't load as much as an elephant, improve your load capacity 
Appraisal for Elephant - You take heavy loads but you are slower than a horse, We need you to be good at speed
In the end, both of them will become weak :)
Remember you are your own competitor

Keep Running!!!

July 14, 2023

Docket Notes - Image Sizes

Docker: Reduce the size of a container

using https://alpinelinux.org can reduce the size and deployment time of a container by up to 40%

Install minimum packages

Reducing Docker Image size

  • Bulkier images take time to get pulled.
  • Increased time to spin up a container using it.
  • Include a .dockerignore file.
  • Reduce unnecessary dependencies.
  • multistage build pattern. Made a binary on the build stage and used it to run the container.

.dockerignore

docker-squash

  • Squashing helps with organizing images in logical layers. Instead of having an image with multiple (in almost all cases) unnecessary layers - we can control the structure of the image.

Keep Exploring!!!


July 12, 2023

Why consulting does not impress me?

  • Short cycles with limited data exposure limit understanding of solutions. 
  • Innovation needs iterations, and scoped work does not allow for a deep dive to solve the problem at scale. 
  • Tech expertise is important, but it is equally important to solve the right problems in the right way. 
  • The perspective of domain and business problems with tech solutions often lacks, and quicker cycles are good, but meaningful and impactful solutions are what matter, not intermittent or stop-gap solutions. 
  • Building a world-class solution requires goal-oriented thinking, not task-based thinking.

Keep exploring!

July 10, 2023

Experience = Domain + Data + Tech + Persistence

I could deeply relate to this post. I believe that having a deep understanding of the data and domain knowledge is essential for producing the best results. I have experienced this firsthand in my work, such as rewriting warranty engines, designing solutions to meet the needs of the business, coming up with approaches for warranty corrections, and supporting on-the-fly changes for different markets.

The relevance and completeness of the solution is the result of production support, familiarity with the tables and columns, and an awareness of any data issues.

In order to achieve the best results, it is important to have a comprehensive understanding of the data and domain knowledge. This requires consistency and perseverance, and it is only through this that the big picture can be seen.

Having a thorough knowledge of the data and domain can be difficult to acquire, but the rewards are worth it. Not only will it help to produce better results, but it will also give you a better understanding of the project as a whole. It is also beneficial to have a good understanding of the different markets and the changes that may be required for them.

Having a deep understanding of the data and domain knowledge is essential for producing the best results. It requires consistency and perseverance, and it is only through this that the big picture can be seen. It is important to have a comprehensive understanding of the data and the domain knowledge, as this will help to produce better results and give you a better understanding of the project as a whole. Additionally, it is beneficial to have a good understanding of the different markets and the changes that may be required for them. With the right knowledge and understanding, you will be able to produce the best results for any project.

Learning's as part of current roles

  • Discuss with both between business and technical audiences.
  • “Hands-on” and “agile” to fix where it breaks
  • Technically hand-holding data scientists, engineers, and architects depending on solution needs. Drive back to first principles
  • Collaborating with colleagues from different cultures and time-zones across the globe

Keep Exploring!!!

July 01, 2023

Finance - Learning

What are payables in finance ?

Payables in finance refer to the amounts owed by a company to its creditors, suppliers, or other entities as a result of purchasing goods or services on credit. These are short-term financial obligations that need to be settled within a specific period, usually within a year. Payables are recorded as liabilities on a company's balance sheet and are also known as "accounts payable" or "trade payables."

In simple terms, payables represent the money a company owes to others for goods or services it has received but not yet paid for.

Does financial planning happen with cash in hand or planned cash and estimate with it

Financial planning typically involves both cash in hand (current cash available) and planned cash (future cash inflows and outflows). It is a comprehensive process that helps individuals and businesses to manage their financial resources effectively, achieve their financial goals, and ensure long-term financial stability.

In financial planning, you start by assessing your current financial situation, including cash in hand, assets, liabilities, and income sources. Then, you set financial goals and create a plan to achieve those goals by estimating future cash inflows (such as income, investments, and other sources) and outflows (such as expenses, debt payments, and savings).

The plan should be flexible and adaptable to changes in your financial situation, goals, and market conditions. Regular monitoring and adjustments are essential to ensure that the plan remains effective and aligned with your financial objectives.

In summary, financial planning involves working with both cash in hand and planned cash, as well as making estimates and projections to create a comprehensive and adaptable plan for achieving financial goals.

What are key factors that influence payables in finance ?

Several key factors influence payables in finance. These factors can affect the amount, timing, and management of payables for a company. Some of the key factors include:

Credit terms: The credit terms offered by suppliers, such as payment due dates and discounts for early payments, can significantly impact the amount and timing of payables. More favorable credit terms can lead to lower payables, while less favorable terms can result in higher payables.

Volume of purchases: The volume of goods or services purchased on credit directly affects the level of payables. Higher purchase volumes lead to higher payables, while lower volumes result in lower payables.

Payment policies: A company's payment policies, such as the frequency and timing of payments to suppliers, can influence the level of payables. Companies that delay payments or negotiate longer payment terms with suppliers may have higher payables, while those that pay promptly or take advantage of early payment discounts may have lower payables.

Economic conditions: Economic conditions, such as interest rates, inflation, and overall market conditions, can impact payables. For example, higher interest rates may encourage companies to pay off their payables more quickly to avoid additional interest costs, while lower interest rates may lead to companies taking longer to pay their payables.

Supplier relationships: The quality of relationships with suppliers can influence payables. Companies with strong supplier relationships may be able to negotiate better credit terms, resulting in lower payables. Conversely, strained relationships may lead to less favorable terms and higher payables.

Industry practices: Industry practices and norms can also affect payables. Some industries may have standard payment terms or practices that influence the level of payables for companies operating within that industry.

Cash flow management: Effective cash flow management can help companies optimize their payables by ensuring that they have sufficient cash to meet their obligations while minimizing the cost of carrying payables. Companies with strong cash flow management may have lower payables, while those with poor cash flow management may struggle to meet their payment obligations, resulting in higher payables.

Understanding and managing these factors can help companies optimize their payables, improve cash flow, and maintain strong relationships with suppliers.

Keep Exploring!!!