AI Projects - Inspirations - Notes - Post 2 (10 Projects)

Project #1 - Music Recommendation

Key Observations

• Free and customizable recommendations
• Dataset - spotify million playlist
• Collaborative filtering model
• Two-stage nearest neighbourhood model

• Popularity, Energy 

Project #2 - LegalEgo

Key Observations

• Contract Review documents
• Automated annotations

• Output labels and annotations

Project #3 - Autosuggest

Key Observations

• Articles that support our claim

Project #4 - Cryptocurrency prediction

Key Observations

Project #5 - Fashion Recommendation

Key Observations

Project #6 - Instaavatar

Key Observations

• Random Avatars

Project #7 - Image to Calories detection

Key Observations

Project #8 - Augmented Image Search

Key Observations

Project #9 - Product search

Key Observation

Project #10 - Sign language

Key Observation

Keep Exploring!!!

AI Projects - Inspirations - Notes - Post 1 (10 Projects)

Project #1 - Generate NFT style images

Key observations

• Generate images
• Convex combination
• Image denoising

• Architecture

Project #2 - Emoji generation by text

Key observations

• Emoji adoption on rising
• Impact of Emoji

• Architecture

Project #3 - Trading by breaking news

Key Observations

• Collect headlines
• Analyze and inform
• Summary and Stock opinion
• Bearish / Bullish

• Keywords, similar articles

Project #4 - Vision powered cooking items

Key Observations

• Use vision to map ingredients to the recipe
• The key thing is vegetables, ingredients detection

Project #5 - Misinformation detection

Key Observations

• Auto detect misinformation
• Confirm or dispute info comparing sources

• Prediction and Explanation of it

Project #6 - Edit audio by editing text

Key Observations

• Audio corrections
• Edit and remove and retain only needed items

Project #7 - Schedule power and reduce consumption

Key Observations

• Reduce emissions / Track by activity
• Retrain every hour
• Hot swap models

Project #8 - Political bias detection

Key Observations

• Informed in fairway
• Classify biased news sources
• Highlight bias in news sources
• Polarity chrome extension to detect biases 
• Highlight portions that attribute to bias

Project #9 _ Photoapp to summarize

Key Observations

• Multiple instances of pics
• Clusters images into sensible scene classes
• Rank pics by the aesthetic quality
• Clustering and quality assessment seamlessly

Project #10 - 911 operator assistant

Key Observations

• Respond faster
• NE, Emergency detection
• Nearest Help
• Google Maps API
• Closest route

Stay Tuned for Part II from 1hr 20 mins

Keep Collecting inspiration!!!

MlOps - Lets Learn it - Day #1

Many thanks to Stanford and MIT for sharing knowledge. In 2000 I used to download PPTs and refer to slides. Now you have a ton of materials plus a lot of distraction. Filtering knowledge vs chaos need time, focus and bookmarking.

Course materials are shared in Link

I wanted to review the first lecture/notes and bookmark my lessons

Notes - Link

Summary from it

Machine learning is an approach to (1) learn  (2) complex  (3) patterns from (4) existing data and use these patterns to make (5) predictions on (6) unseen data.

• Learn: DB has explicit relationships but ML learns relationships
• Complex: Across attributes ML finds relationships
• Patterns: Influences, categories, segments ML finds
• Existing data: Learn from data, improve on ongoing data collection. 
• Predictions: Use the learnt knowledge to apply for incoming data

Use cases distributions, Top 3 outside costs

• Customer insights
• Improve experience
• Retain Customers

Data Hierarchy - Data - OLTP - OLAP - ML - AI

Difference between Latency and throughput

Keep Exploring!!!

Anamoly / Outlier Detection packages

 

Ref - Link

• Pyod - Link
• Timeseries anomaly detection using an Autoencoder
• Autoencoder For Anomaly Detection Using Tensorflow Keras
• keras-anomaly-detection
• SRCNN Anamoly Detection

Keep Exploring!!!

ImageHash package - simplify hash

Python library to hash by different methods. Useful to build indexes / similarity search.

The different hashing techniques are

• Average hashing  - crushes the image into a grayscale 8x8 image and sets the 64 bits in the hash based on whether the pixel's value is greater than the average color for the image.
• Perceptual hashing - use a discrete cosine transform (DCT) and compares based on frequencies rather than color values
• Difference hashing - gradient hash, calculate the difference for each of the pixel and compares the difference with the average differences.
• Wavelet hashing - works in the frequency domain as pHash but it uses DWT instead of DCT.
• HSV color hashing (colorhash)
• Crop-resistant hashing

Ref - Link1, Link2

Ref - Link

Package Installation

File Upload

Create Hash Methods

Depending on usecases need to evaluate which suits the need or a combination of techniques.

Keep Exploring!!!

Learning moments

#Learningmoments - Getting things done, Delivering functional code on time is a key aspect of development. Option #1 - If find it already solved, I lookup #StackOverflow / #blogs and #integrate it works. After It is integrated, I move on to other #pieces of the problem. Option #2 - Some solutions even if we evaluate solutions in StackOverflow / blogs do not completely address our use case scenarios. After initial exploration, whiteboard, get back to #basics, get the building blocks again and find #optimal solution. Both #1 and #2 are part of learning. Option #2 takes time but also gives you deep clarity and analysis. Depends on #interest, #timelines, and team setup to deep dive. #Learning / incremental experience adds more perspective as we solve diverse use cases. #experience #learningeveryday #perspectives

The cycle of productivity - focus - inspirations

• Money is the outcome of knowledge
• Knowledge is the outcome of implemented ideas
• Ideas are the outcome of focused domain, data, implementation
• Ideas are the outcome of focused learning
• Focus is the outcome of the purposeful search for inspiration
• Inspiration is the outcome of curiosity and consistently looking out for great ideas
• Everything connects, habits emerge from small incremental motivations...

Reading / Filtering / Experimenting

Inspirations take time to spot

• Reading blogs
• Simple code flows. Well articulated
• Code runs in mind when you see it
• Constantly lookout to get/bookmark quality posts
• Share your views/perspectives big or small issues, It's ok write it up

Interesting Note

To get expertise it needs

Since we work from time to time on different problems with familiarity vs deep dive vs practice vs forgetful memory everything impacts. Tech landscape changes very often so again needs to be updated.

From link

• Start with the smallest working example
• Try to modify it and watch the results 
• Read or watch someone explain the technology
• Repeat steps as often as necessary

100 productivity tips

𝐈𝐧𝐭𝐫𝐚𝐩𝐫𝐞𝐧𝐞𝐮𝐫'𝐬 10 𝐂𝐨𝐦𝐦𝐚𝐧𝐝𝐦𝐞𝐧𝐭𝐬 𝐛𝐲 𝐆𝐢𝐟𝐟𝐨𝐫𝐝 𝐏𝐢𝐧𝐜𝐡𝐨𝐭

🔸 Come to work each day willing to be fired

🔸 Circumvent any orders aimed at stopping your dream

🔸 Do any job needed to make your project work, regardless of your job description

🔸 Find people to help you

🔸 Follow your intuition about the people you choose and work only with the best

🔸 Work underground as long as you can --publicity triggers the corporate immune system

🔸 Never bet on a race unless you're running in it

🔸 Remember, it's easier to ask for forgiveness than for permission

🔸 Be true to your goals but be realistic about the ways to achieve them

🔸 Honor your sponsors

--Gifford Pinchot, Intrapreneuring, Harper & Row, 1985

Keep thinking!!!

Flask - Image Attachment - Samples - API

 

More Read - Deploying a Flask app on AWS Lambda with Zappa

Keep Exploring!!!!

Segmentation Notes

Many thanks to this article. Segmentation worked fine on cars / humans. If we could retrain this for larger image would be great

colab notebook link

Demo code link, article2

Sample output

Dataset prep script - Link1

Model codes - Link1, Link2

Keep Exploring!!!!

Computer vision challenges - Data Challenges / Updates

• Dataset challenge - Easier to build age/gender with kaggle. If you need to build age gender for one country you can't find the dataset easily
• Implementation challenges - Segmentation models in the real world do not work 100% perfectly. Github code to expected accuracy takes time
• Post-processing - Color correction, noise detection, depth estimation multiple techniques we need to have to have a final smoothness
• Maintenance / Update issues - mediapipe was released in 2020, The architecture/datasets are not shared in the general forum. There is no updated model at this point

The learning list consists of

• Vision models awareness
• Vision image processing techniques
• Domain related aspects
• MVP product knowledge
• Making it work at scale 
• Research Landscape
• Data collection / techniques / models

Keep Exploring!!!

Debugging / Code Practical Views

• Modularize and integrate it easier to debug
• Clarity of code changes vs perception of change, Know every key line of change, Why and how it works
• Easier to plug and test smaller portions of code as needed
• Documentation at method level on input/output
• Comment key lines of function in code (the core of logic), Keep versions of old key lines commented as needed. 80% of key logic may be in 2 lines of a function :)
• Logs to capture
• Spot the block that errors / Enable / disable the key lines / retain prior changes to spot low-level issues

Knowing purpose makes it easier to spot the line!!!

Keep learning code through a product lens!!