Learning Notes - Optimization

Key Notes

• Three parameters for Linear problems
• Decision variables (Number of quantities to decide)
• Objective (Minimize / Maximize profit / time)
• Constraints (Time / Resources)
• Identify these three parameters for your problem
• Leverage existing Packages

How it solves

• Decision variable multiplied by cost, subject to the constraint
• Different modeling frameworks

Fundamentals (Link)

• Convex optimization involves minimizing a convex objective function
• Linear programming is a special case of convex optimization where the objective function is linear and the constraints consist of linear equalities and inequalities
• Linear programming is a special case of convex programming, in which the objective function is a linear

Link, Link1

• Optimization is when you search for variables that attain a global maximum or minimum of some function
• Convex optimization is a subset of optimization where the functions you work with are "convex" which just means "bowl shaped". This makes the search for maxima and minima easier since you can just " walk " on the surface of the bowl in the direction with the greatest slope to get there.

Keep Thinking!!!

Learning Notes - Coresight Classroom: Retail Business Models Evolution in China Under COVID-19

This was a useful webinar. It showed examples in China in new forms of business, innovation, going digital, patterns of shopping.

Trend #1 - Livestreaming for consumer experience

Most of the businesses have moved online integrating a live stream shopping experience. This is more of connected / virtual and contactless shopping

• Starbucks - Contactless delivery in China
• Most innovation during crisis - Restaurants
• Whatsapp for buying/selling
• Whatsapp groups 

Retailers

                                                                       Shopping Malls

Luxury brands going digital

•  Luxury brands opened flagship stores online

Restaurants going live for ordering/enhance customer confidence

• Restaurant to see online and buy through platform

Trend #2 - Older users going Digital

Trend #3 - Local brands emerged as preferred suppliers / Conservative / Sentiment to support local industries

Trend #4 - Offerings for stay-home situations

Trend #5 - Consolidate CRM instead of fancy marketing / New Post COVID World

Keep Thinking!!!

Invisible Risk

A quick read of two articles provides good insights on mediums of contracting risk and areas applicable in different domains.

Article1 - Link
Contracting coronavirus

• High touch surfaces like door handles, elevator buttons 
• Cough - 3,000 droplets and droplets travels at 50 miles per hour, Washroom, Foodcourt
• Sneeze - 30,000 droplets, up to 200 miles per hour
• Breath - releases 50 - 5000 droplets

Successful Infection = Exposure to Virus x Time

Article2 - Link
Mitigate / Safety Precautions
Retail

• High touch surfaces - doors, shopping carts, self-checkout counters, billing machine
• Conversations - Customer to Customer, Customer to Cashier, Customer to Shopper associates

Manufacturing

• High touch surfaces - CNC machines, parts, assembly, inspection
• Conversations - Coordination between workers, Inspection, Assembly

Warehouses

• High touch surfaces - RFID readers, cartons, products, packaged items
• Conversations - Coordination between workers, shipment, returns

Reality is harder than this. Hope this learning creates more awareness on health and safety.

Reality Check - Link

Keep Thinking!!!

Retail Recovery Early Signs

A lot of innovative ideas/experiments I have witnessed in discussions/webinars to provide Safe Retailing and Recovery path for offline Retailers

My Personal Observations

• Never stepped out since March 23rd
• Essential supplies Milk / Curd / Vegetables through Supr (Digital Shopping)
• Groceries purchased via swiggy  (Digital Shopping)
• Working style, Learn / Code @ home has been rewarding
• The pattern of corona hasn't seen a decline, in fact, it feels scary, How do I prepare to get into a social circle

What Retail has seen in Feb / March / April

• Store closure
• Furlough
• Loss of jobs/possibilities of bankruptcy

Since May, Stores across the world are preparing for the new normal. Retail community is working together to get back to safe Retailing

Key Observations

• Shoppers go digital and provide live support/adoption of live chat/video streams for a personalized experience
• More offline store go digital leveraging platforms like Shopify
• Exploring Contactless shopping with Physical Distancing, Signs, Kerb Side pickup, Appointments

AI Opportunities

• A lot of video analytics opportunities for Compliance, alerting, contactless shopping, chatbots

The Future
There are a lot of challenges in terms of operating in a new environment. There are different segments of customers post COVID scenario. The new shopping patterns, panic buying, health-conscious buying everything will challenge the demand forecasts during the normal operating environment. The store layout also will change, the way items are arranged, AISLE layout, and planogram.

Realistically we can look at

• Going digital
• Restart in a Staggered manner
• Limit Focus on certain segments of products

Apparel / Luxury shopping would need a few quarters to rebound to normal activity. The recovery also depends on economic factors, local government support, economy, and population mix. It will take a few quarters for consumers to get used to the new normal.

References
Link1
Link2

Keep Thinking!!!

Error - 'protoc' is not recognized as an internal or external command - Windows 10

Happy Learning!!!

Predicting the Future: Retail's New World - The Parker Avery Group

• Historical models will be compromised. Allocation, Replenishment and Demand Forecasting
• Learn - Recover - Evolve
• Preparing ourself for next time, learn to recover from the current situation

Learning

• Store closures 
• A significant difference in sales patterns
• 2018 vs 2019 vs 2020
• Execution systems, Replenishment, Forecasting, Automated purchase orders, time-series approaches 
• Forecasts systems need intervention 
• Analytical systems haven't seen such situations

Recover

• Reassess the demand
• Validate Forecasting Numbers / Adjust according to the market situation
• Sales mix will look very different at category/class/subclass level
• Trust your gut feeling
• Having a cross-functional team that understands the demand/operations/foot traffic. All functions planning, fulfillment, store managers work together. Retail community learn from each other
• Test and Learn for short term
• Forecast for shorter timelines and estimate the accuracy of models 
• How long reduced sales will continue depends on several factors / economic conditions
• Ensure data is in the best shape to capture the trends
• Identify signals in short / medium / long term
• Understand sales mix identified online and bring it back to offline stores
• Do not open all stores at once

Evolve

• Go beyond time series, Decision Trees / Deep Learning to include hierarchy/location attributes / COVID parameters/impact - Look at short term / long term forecasts. Supplement with newer features
• Supply Chain - Ensure inventory is in place during the crisis situations. 

Time

• How long mitigation process will take?
• Shorter the better
• Start with easy mitigation strategies
• Video - Faster responses / Right time to invest
• Make existing systems smarter (Intuitive)
• Sales are going to be down, the sales mix will be different. Have a rolling sales mix

Learn from current disruption to prepare for the next disruption

Keep Thinking!!!

IHL Retail Webinar

Speaker - Greg Buzek (Greg Buzek is the Founder and President of IHL Group)

Data Perspectives

• Don't sensationalize data
• Look data in the context of region / local culture

Recovery Path 

• Horrible Q1 / Q2 until 2021
• Recovery for each region has several dependent factors like 
• Recovery Models / Stimulus packages / Gas prices
• Stores may reopen in Chunks 10%, 20% in incremental levels

Post-Covid

• Increased Mobile Payments / DBT
• Digital Transactions Adoption
• New store formats 
• Rise of frictionless stores
• Autonomous Retail Environment

I disagree a bit here, RFID is an expensive investment compared to AI / ML solutions. In this current situation, RFID requires high initial investment costs. This makes it the right spot for AI / ML-based solutions

Keep Thinking!!!

Notes - Object Tracking in Videos: Approaches and Issues

paper - Link

Learning is the outcome of failed experiments, solving the problem in your own way, finding a solution based on learned, and newer ideas.

Key Notes
Flow Detection -> Classification -> Tracking -> Action Recognition

Tracking Approaches

• Point Tracking - Track based on points tracked in detected Objects
• Silhouette Tracking - Track based on appearance region in each frame
• Appearance Tracking - Appearances in consecutive frames

Key Techniques

• Dominant color separation technique
• Upper body dominant color
• Lower body dominant color
• Kalman filter
• Haar-like features classifier
• Similarity scores are computed for these subparts / Visible parts
• Centroid based approach
• Mean Shift Algorithm
• Silhouette trackers - shape matching and contour tracking

We can also experiment transfer learning-based feature extractor, clustering as well to group / classify them into groups and detect accordingly

Keep Thinking!!!

Weekend Lectures - Andrej Karpathy: Tesla Autopilot and Multi-Task Learning for Perception and Prediction

Key Lessons
Single Task

• For Lane tasks to do are - cars detection, multiple car types, take complete viewpoint, occluded cars
• Architecture, Loss function, Object detection techniques
• Train - Test - Fix until you get the required accuracy

Multi-Tasks

• For a moving vehicle, the tasks to compute are Static Objects, road signs, overhead signs, Traffic lights, Lane Lines, Road markings, curbs, crosswalks, environment tags
• All these inputs will be worked on simultaneously
• Single tasks, Subtasks of moving objects
• For every vehicle - vehicle type, lights, indicators, blinkers. All are independent predictions

Key Learnings

• Architectural considerations
• Loss function considerations
• Training dynamics
• Team workflow
• Feature sharing at which layer

Architecture choices

• Moving objects
• Static objects
• Signs
• Traffic lights
• Decide frame rate according to tasks
• Feature sharing between tasks
• Tasks to create features

Paper References
Auto DeepLab
Which tasks should be learned together in multi task learning

Integrating multiple views

• 8 cameras inside the vehicle
• Different viewpoints of each camera
• Some of the layers can be shared as edges/shape can be similar
• RNN sharing features from different frames
• Static parts / Moving parts / path prediction
• A lot of domain-driven optimization
• Sampling tasks / sub-sampling networks / Sharing features

Loss functions

• Each task will fill into same loss and backpropagate
• Panoptic feature Pyramid networks 
• Include both Object detection and Semantic segmentation
• Large grid search over task weights to find best mean average precision

Training Dynamics

• Data Distribution
• Over Sampling 
• Within task oversampling (Liked it)
• Semi balanced batches
• Data Engine for each task and dedicated instances
• Single task - Early Stopping when validation loss is lower

Session #2 - Deep learning applications: training a multi task classifier

• Fetch Product, Price, Feature, Time <-> Price mapping, Multiple prices over time
• 90K brands
• Single task vs Multi-task classifier
• Leverage current Keras custom loss generator

Multi-task

• Leverage a few layers common for multiple class
• Classify 2 labels vs n labels
• Right product @ Right price @ Right time = Best price @ available margins @ Right time

Example of Multi-Task

• Vehicle Classification + Vehicle Color Detection
• Vehicle Classification + Type detection

Keep Thinking!!!