Smartphone / Social Media Issues

There are a lot of issues due to cheap internet smartphone / social media, The primary motive seems to have failed

• Targeted marketing to influence people
• Change from Morality to Majority Opinion is right
• Crowd following mentality
• Drive more consumeristic / purchase behavior
• Focus more on lifestyle/luxury than principles/priorities
• Low or no importance to education
• Setting up wrong examples
• Addiction to games/media / fake gurus
• Commercialize everything, Take away all the time
• In a way they have created isolated hotspots, depressed, addicted to smartphones. With cheap internet, they have only screwed their lives further. Targeted apps for each age group. Drive based on their emotional, behavioral needs. 

Keep Questioning!!!

Interesting Research paper Read - Gender and Race Preferences in Hiring in the Age of Diversity Goals: Evidence from Silicon Valley Tech Firms

Paper - Link

Key Insights

• Women are 9-10% more likely to receive a callback compared to men, 
• Whereas Black Hispanic and Asian applicants are 8-13% less likely to receive a callback compared to White applicants

Key Notes

Studying hiring discrimination at the intersection of race and gender, giving primacy to both

How hiring discrimination, in particular, leads to occupational segregation.

Experiment #1 - Send Fictitious resumes with randomized white-sounding and black-sounding names to potential employers for different types of occupations and consistent discrimination against African Americans across occupations (Bertrand and Mullainathan 2004).

Insights

Statistical discrimination - Employer who imperfectly observes an applicant's quality and productivity resorts to group-level averages to make inferences about the individual, which may lead to discrimination

Taste-based - Employers may have a prejudiced taste and animus towards a particular group, leading to discrimination (Becker 1971).

Discriminatory phenomenon -  female discrimination in male-dominated occupations and male discrimination in female-dominated occupations

ML Approach

Keep Thinking!!!

Personal Datawarehouses

We need to have the ability to claim our personal data / use it to trade without PII. Everything now is paid Gmail / google photos/youtube. The end-user data is used without any benefits for End-user.

• Reclaim your Google Data
• Reclaim your Social Media Data
• Reclaim your Amazon Data
• Reclaim your Location Data

Hope there is some value for user data.

Keep Thinking!!!

Session - ODSC East Talk - Challenges of machine learning development

Code - Link

Key Notes

• Automated and Scalable Infra for ML 

ML Automation Steps

• Data Pipeline
• ML Team
• Production / Deploy / Feedback Mechanism

Key Challenge is integration of all stages of development

• Reproducibility via docker
• Scaling via Kubernetes

Reproducibility

• Share insights
• Deploy Code

Automation process

• Data pipeline
• Feature generation

Production

• Monitoring
• Logging
• Packaging

End to End workflow of Development to Production

Infra 

• Deploy
• Monitor
• Train
• Scale it on cloud

Data Pipeline process

Keep Thinking!!!

Weekend Reads - Advanced Models for Computer Vision

Key Notes

What Classifier will Miss - Human-level scene understanding

• Parsing the scene
• The angle of Bicycle (Pose, Relative pose)
• Person on Bicycle
• Closer Inspection

Tasks

• Object Detection
• Pose Estimation
• Accuracy vs Efficiency of Models

CNN as Deep Learning Puzzle

Input-Output Node, Loss Computation and Backprop

Classification - sparse description of the image

Object Detection

• Multi-task problem
• Classification & Localisation
• Object, Location, Bounding box
• Dataset, Samples, List of Objects, Labels, Bbox for each object

Predict BBOX Coordinates

• Continuous Output
• Minimize mse of samples
• Regression for bbox prediction
• The first part is the classification
• The Second Step is regression

Faster RCNN

• Two-Stage Detector
• Good Candidate BBOX
• Refine through Regression
• Discretize bbox space
• Anchor points distributed
• Candidate boxes of different scale and ratio
• n candidates per anchor
• Is there an object or not in the box
• Refine through regression
• We cannot backdrop on parameters of bbox (Spatial Transformer Networks)

One Stage Detector - Train end to end

• Employ Hard negative mining

Retinanet uses Focal Loss (The loss function is just a mathematical way of saying how far off a guess is from the real value of a data point.). It puts more weight on the objects that were hard to classify and decreases the impact on easy correct predictions

Semantic Segmentation

• Pooling - reduce the resolution of feature maps
• Upsample based on the nearest neighbor approach

U-Net

• Segmenting medical images
• Input Image -> Convolution -> RELU pooling
• Encoder - Similar to Image Classifier
• Upsampling through Decoder for same resolution output
• Upsampling - blobby feature map
• For every location distribution over classes
• Cross Entropy (Avg Over all Locations)

Keep Thinking!!!

Next Paper Read - Docker, RDBMS to ML

Paper #1 - An introduction to Docker for reproducible research

Key Notes

Docker provides a binary image in which all the software has already been installed, configured and tested

Technical Issues in Software Deployment

• Software Dependency Hell
• Imprecise documentation

Docker Features

• Performing Linux container (LXC) based operating system (OS) level virtualization
• Portable deployment of containers across platforms component reuse
• Versioning of container images
• Docker images share the Linux kernel with the host machine
• Sharing the Linux kernel makes Docker much more lightweight and higher performing than complete virtual machines

Components

• Dockerfiles provide a simple script (similar to a Makefile) that defines exactly how to build up the image
• Docker also supports Automated Builds through the Docker Hub (hub.docker.com).

Paper #2 - The Relational Data Borg is Learning

Key Notes

• RDBMS in Data Science
• Widespread need for efficient data processing
• Process beyond classical database workloads
• From the Survey 65% data is Relational. Retail has maximum structured data :)

Automated Feature Learning Approach

Key Features for Retail Stores

• Items in stores
• Store information
• Demographics for areas around the stores
• Inventory units for items in stores on particular dates
• Weather Information

Queries based on Filters

• Feature extraction query that joins these relations on keys for dates, locations, zipcode, and items
• LMFAO (Layered Multiple Functional Aggregates Optimisation) 
• PCA over relational data

Insights

• Running aggregates over days, weeks, months; min, max, average, median aggregates, or aggregates over many-to-many relationships and categorical attributes

ML Tasks

• One-hot encoded
• Categorical attributes
• New database workload motivated by a machine learning application
• Similar aggregates are derived for k-means clustering

(Iterative Functional Aggregate Queries) Framework

• IFAQ can automatically synthesise and optimise aggregates from ML+DB workloads

Key Insights / Lessons

• Turn the learning problem into a database problem.
• Exploit the problem structure to lower the complexity.
• Generate optimised code to lower the constant factors

There is no Data Science without Database - RDBMS :) :)

Happy Learning!!!

Working on Weekends

 This tweet sync up with my opinions 

Unpopular opinion: Weekdays/weekend split isn't for everyone. It forces you to:

• work in low-energy days (just bc it's a week day)
• recharge on preset Sat/Sun days (you can't schedule when you'll need rest)
• rest two days in a row (when separating the two could do wonders)

— Ali Mese (@meseali) November 1, 2020

There are no fixed learning hours, moments. It is actually when you feel an optimal working environment that makes you more productive, the result is maximum learning impact.

For me weekdays build the list of To-do lists and weekends help me to find the right resources, examples to figure out and complete the list. There are learning moments on weekdays but It again depends on meetings/calls and the focus time that you do. 

This picture summarizes the perspective

Another good read, Fixing priorities, and focus. This is a great post in those lines.

Keep Thinking!!!

5 Reasons Why you need to build your Inhouse Vision Solutions

• Opensource models may not work 100% for all scenarios, You need to customize perform transfer learning for your environment
• Buying black box models/products may not help and it will cascade into substantial additional investments for maintenance and support
• It is an evolving product with different situations, locations, we need to invest and build expertise, Starting with open source and customizing it will help in building stable solution in the long term
• Vision needs to be looked up as additional insights from your existing CCTV data, Alternative solutions for RFID based solutions
• Vision needs to work with transactional data. Both need to be seen together to get insights and translate into key actions

Keep Thinking!!!

Interesting Use Case - Sports Analytics

Fantastic Computer Vision Use Case

• Person Tracking
• Action Recognition
• Heatmaps

Computed Insights

• Player Tracking
• Speed
• Positioning
• Defense / Pass details
• Analyze the opponents

Keep Thinking!!!

Casual Impact - Paper Read

Paper #1 - Link 

Key Points

• The causal impact of a treatment is The difference between the observed value of the response and the (unobserved) value that would have been obtained under the alternative treatment
• Data #1 -  The first is the time-series behavior of the response itself, prior to the intervention.
• Data #2 - The second is the behavior of other time series that were predictive of the target series prior to the intervention
• This selection is done on the pre-treatment portion. Value for predicting the counterfactual lies in their post-treatment behavior

Summary of Steps Link 

• Fitting a Bayesian structural time series model to observed data 
• Predict with Intervention
• Predict without Intervention 

Paper #2 - Causal Impact Analysis for App Releases in Google Play

• Causal impact analysis uses a control set: a set of unaffected data vectors
• Experiment using different control sets in the study
• The agreement is defined as (YY+NN)/total
• YY indicates a significant change as detected on both datasets
• NN indicates no significant change detected on both datasets

Keep Thinking!!!