via xkcd @ https://xkcd.com/2054/

Considerations Choosing the Best Workflow Management System

• Built in Scheduler

• Easily deployable on AWS

Nature of Data Ingestion [Best effort Basis]

1. Receive large amounts of data regularly and sporadically, we’ve decided to perform this action once an hour, ensuring we don’t miss out on any important data.

2. A delay in data delivery because of the source platform’s poor performance

3. Data pipeline that hits an outside service to get some data and that service happens to be down (outage), or maybe the format of the data changes

4. Fluctuating amount of data being consumed and outputted from the source

5. Frequency of on demand polling the data source is conflicting duplicating scheduled polling.

6. Data gets clogged up at the slow step; this leads to a bottleneck in the pipeline.

Why use Workflow Management System? [Frameworks for Data Pipelines: ETL workflow framework, with a scheduler that would trigger the appropriate tasks it is programmed to]

1. No Loops, DAGs Only: Loops in pipelines are bad. There is a reason most (if not all) frameworks available enforce DAGs for pipeline building: cycles introduce unnecessary complexity and make it very difficult to reason about your system.

2. Avoid second order effects of writing custom pipeline from scratch, all pipelines require some custom effort; the real question is how much you actually need to do, and what framework will support you best.

3. Accomodate changing requirements for data pipelines (large or small) with backword comptability and flexibility for fast iterations i.e first-class support for dataflow allowing flexiblibility of pipeline to be designed to abstract the database logic into a separate set of tasks from the core data transformations. This would minimize the impact of changing those tasks (e.g. changing the underlying data warehouse)

4. Managing complex workflows via a proven Scheduler. When trying to run periodic tasks you decide to write an algorithm that takes the output of one job and uses it as the input for another job. This task depends on preceding jobs to be completed successfully, because they have to run in chronological order. If the first task didn’t run properly, it could create errors later on. Your scheduler should be able to handle those kinds of situations, without requiring you to be constantly on the lookout for bugs and errors while wading knee-deep into your code and its dependencies. You might fall into the trap of building these schedulers yourself. However, this is probably not the best idea, because it’s likely you will encounter challenges that can be easily overcome by automation.

5. Metrics, Metadata, & Logging: keeping track of data pipeline metrics and quickly diagnosing issues that you were previously unaware of [Grafana for metrics, Logstash for logging and sentry for errors]

6. Fallback design for unexpected external data sources scenarios: sample this scenario in your data pipeline that hits an outside service to get some data and that service happens to be down, or maybe the format of the data changes. Your pipeline needs to know how to handle cases like this to prevent errors from taking down the system. Mechanisms such as retries, error handling, fallback alternate data sources queueing can really solve this problem

7. Scaling Out in the face of bottlenecks: A delay in data delivery is a critical error in pipeline development, especially when that pipeline is used in a production environment. Cluster auto-scaling and web server load balancing are life savers: use them. A system of scale needs to be planned out and (preferably, as to avoid another aspect of maintenance) automatically enabled. One way of doing this consists of placing checks which are aware of the amount of data queued up to be processed for that step and scaling the amount of resources accordingly.

References

https://medium.com/the-prefect-blog/pipeline-pitfalls-57fe558cd76

https://otonomo.io/blog/luigi-airflow-pinball-and-chronos-comparing-workflow-management-systems/

https://kapernikov.com/a-comparison-of-data-processing-frameworks/

https://stories.dask.org/en/latest/prefect-workflows.html

https://medium.com/snaptravel/building-a-data-pipeline-to-handle-50m-events-day-with-a-startup-budget-part-1-636399bd00aa

https://towardsdatascience.com/data-science-for-startups-data-pipelines-786f6746a59a

It’s like supplying guns & ammunition to both consumers and producers on demand for consumption and creation with instant feedback loop cycles

• Consumers (audience) side: Tik Tok uses AI to calculate what content will be the best fit for the consumer. When consumer use the app, consumers even do not need to put their preference. AI will automatically get a sizeable database of the consumers’ behaviours.

  “It’s live a television on your palm without the decision overload of deciding what to watch (think no fighting for the remote, your own personalised television) feeding on your subconscious attention span (i.e social network dopamine sprinkled on top) and micro personalised to your tastes and liking based on watch time analytics gathered over your historical behaviour (past episodes of using the Tiktok App)”

  

VS

• Producers (content creators) side: Tik Tok will help content creators build the virtual videos, by editing with music, filters and other enhancement. An example is how Tik Tok using technology to create controlled rain in this video: 

“It’s video filters using advanced AI algorithms to create deep fakes, morphed videos and unreal scenarios. Think how recently Virtual backgrounds(Zoom), Beauty Filter enhancers(Beauty plus app) and Replacing Faces in Videos(Face App) have gone mainstream in the year 2020 without much objection but only fan fare of magnitudes unseen even in Social Media network adoption benchmarks.”

Reference this sampled video

• https://mc.ai/understanding-tiktok-how-ai-changes-platform-design-and-economics/

SAMPLE INCIDENT

“Purely digital personas controlled by Russian-intelligence agents.” going by the name

Here is Post-Mortem cum diagnosis cum apology by CounterPost on sourcing content from this fictitious Sophia Managal( allegedly controled by Russian Agenceis - FBI)

In depth Coverage : https://www.theatlantic.com/ideas/archive/2020/09/future-propaganda-will-be-computer-generated/616400/

mladdict.com(browser simulator) - basic math concepts for Linear Regression, Neural Network, Recurrent Network and Q learning agent

seeing theory(online book) - visual intro to probability & statistics (intuitive explanations for Basic Probability: Compound Probability, Probability Distributions, Frequentist Inference, Bayesian Inference, Regression Analysis)

r2d3.us (visual explanations) - statistical thinking with interactive design (recommended: Decision Tree Part, Bias and Variance, Design in a World where Machines are Learning)

Bonus: Explained Visually (EV)

Reference:

https://towardsdatascience.com/learn-machine-learning-concepts-interactively-6c3f64518da2

Medium [Follow the People]

https://medium.com/@parulnith

SELF-LEARNING: Online Resources

Learning Pathways: AI Expert Roadmap, Daniel Bourke’s , Daniel Formoso’s

AI Basics: ML Crash Course, Elements of AI

Python: Geek4Geeks(basics), Corey Schafer, AutomatingTheBoringStuff

Free Courses: fast.ai (I, II), CS50-Harvard, ML-Stanford, mlcourse,

Paid Courses: deeplearning, ZerotoMastery, AppliedAICourse

• Linear Algebra: 3Blue1Brown, MIT(OCW), Rachel Thomas

• Statistics & Probability: MIT(edX), KhanAcad

• Calculus: 3Blue1Brown

Coding Platforms: DataCamp, DataQuest

Bootcamps: FullstackDL,

Books: 100PgML, MLEngineering, MLProblems, Scikit-TensorF

Discover: MadewithML, AIUseCases,

SIGNALLING: Use Online Credentials

Earn Credentials: AI/ML/Data Science (Completely Online):

• (i) Universities(Masters Degrees): Stanford, ImperialCollegeLondon, Texas

• (ii) Universities(Certificates): MITx, Stanford,

• (ii) Online Platforms: Kaggle (Become GrandMaster), Workera, Udacity(Nanodegree: ML Engineer)

Earn Credentials: AI Product Manager (Completely Online):

• (i) OnlinePlatforms: Udacity(Nanodegree: AI Product Manager), Udemy, Udacity(Nanodegree: Data Product Manager)

• (ii)

• (iii)

Earn Credentials: MS in Analytics (Completely Online):

• (i) Universities(Masters Degrees): Georgia Tech(edX),

• (ii) Online Platforms:

  

ROADMAP(s) FOR ML/AI: Pathways

1. [Youtube] Daniel Bourke’s (2.5Hr) walkthrough of a roadMap for learning ML/AI

2. [Notion] Daniel Bourke’s Notion List(41 resources) for learning ML/AI

3. [Map] Daniel Bourke’s playground with a mindmap for ML/AI terrain

4. [Github] Daniel Bourke’s summary of links for his ML/AI mindmap research

5. [Github] Daniel Formoso’s Cheatsheet/Mindmap on the terrain of ML/AI

6. [Github] Daniel Formoso’s Mindmap for DL/AI

7. [Github] AI Expert Roadmap

WHOM TO FOLLOW IN ML/AI: Twitter

1. Industry Experts: Cassie Kozyrkov

2. Proven Instructors: Jeremy Fast.ai, Andrew Ng, Daniel Bourke (Ex-Uber, independent ML/AI Instructor)

Evergreen Resources: Technical Research Papers in ML/AI

• https://arxiv.org/

• https://madewithml.com/

• https://paperswithcode.com/

• AWS Sagemaker for Deployment

Non-related Bonus Content: Architectural 

1. The Death of Microservices

2. Applied AI/ML Vs Theoretical AI/ML

“A flood of derivative articles generated from originals (using AI like GPT-3) will become weapons of mass destruction for humanity by submerging the facts(quasi-truths) with fake news(disinformation)”

This is the threat and this is the challenge

INCIDENT 1:

“Purely digital personas controlled by Russian-intelligence agents.” going by the name

Here is Post-Mortem cum diagnosis cum apology by CounterPost on sourcing content from this fictitious Sophia Mangal( allegedly controlled by Russian Agencies - as per FBI)

In depth apology : https://www.counterpunch.org/2018/01/05/ghosts-in-the-propaganda-machine/

“AI-generated content will continue to become more sophisticated, and it will be increasingly difficult to differentiate it from the content that is created by humans.”

GPT-3 can put words together based on probabilities, it can prompt whenever the author is stuck to express a concept on how it would answer a question.

“The AI’s output tends to ramble a bit and can become somewhat nonsensical if it goes on for too long, but this is a problem humans face as well; it’s why freshman philosophy students have TAs and publications have editors. But given a prompt, GPT-3 can produce any number of unique takes, which a person can quickly and easily polish and post.”

In the past, propaganda needed human hands to write it. While the AI output requires some human editing, it generally produces enough useful material that a manipulator’s job would be much easier, and the content likely significantly harder to detect.

“But the most interesting question that it raises is how people will react and adapt to unreality. If you knew that vast quantities of text on the internet -comments, tweets, articles -was written by fake people, would you care?”

Establishing content provenance?

Notes on possible pathways…

1. It is going to be a matter of concern for readers on internet and that is potentially where blockchain and bitcoin ledger will help in creating original authored content tied to a verified identity locking it in time.

2. There is a slightly middle option: a platform more like dating, or WeChat; less yelling into the void, for others to hear; more double-opt-in chatrooms. That could get weird fast: Ad-driven for low income, paid for the wealthy. Confirmation bubbles is a side effect of this pathway.

3. What has been the response of humanity to previous artificial curation takes like algorithmic recommendations to shopping(amazon) ,music(spotify) or movies( netflix) instead of social recommendations from real friends? The masses have willingly submitted themselves showed indifference to these algorithmic (artificial) curations. Well they are expected to follow suit for fake articles, fake movies, fake music and even fake books (I see a problem with the word….fake…..rather “synthetic” or “artificially” produced articles, movies, music and even books will enter our day to day lives.)

This newsletter piece was summarised from the following sources

1. In depth Analysis: https://www.theatlantic.com/ideas/archive/2020/09/future-propaganda-will-be-computer-generated/616400/

2. 3. The apology written after investigation to it’s readers https://www.counterpunch.org/2018/01/05/ghosts-in-the-propaganda-machine/

   4. The coming age of “fake movies” or “AI generated Movies”

Some Machine Learning experts enjoy making fun of people afraid of Skynet. However, this mockery distracts people from the real issues. 

Every introduction to Machine Learning should include an introduction to AI safety. Readers will forget about Skynet only if Machine Learning experts stop mentioning it, and start talking about the real risks.

Watch the video on AI Safety :

The risk of a super intelligent agent having values not aligned with human values is not something you can laugh at

Elon Musk on safety against AI

Index of Questions that can be answered by AI broadly and are being applied in various sectors. Broadly AI helps answer 5 problem areas

• Is this A or B ?

• Is this Weird ?

• How much ? How many ?

• How is this organised ?

• What should I do now ?

  Machine learning is a large field of study that overlaps with and inherits ideas from many related fields such as artificial intelligence.

  The focus of the field is learning, that is, acquiring skills or knowledge from experience. Most commonly, this means synthesizing useful concepts from historical data.

  As such, there are many different types of learning that you may encounter as a practitioner in the field of machine learning: from whole fields of study to specific techniques.

• Fields of study, such as supervised, unsupervised, and reinforcement learning.

• Hybrid types of learning, such as semi-supervised and self-supervised learning.

• Broad techniques, such as active, online, and transfer learning.

1. Is this A or B ? Classification Algorithms

e.g. 

2. Is this Weird ? Minority in Voluminous Data - Anomaly Detection Algorithms

Define a single class(large pool of good data) - universe of acceptance vs bad

e.g. Fraud transactions

3. How much ? How many ?  - Regression Algorithms 

e.g. Temperature, Climate [Predict Numbers]

4. How is this organised ? - Clustering Algorithms 

e.g. Customer Personas, Netflix categories, Spotify playlists

5. What should I do now ? - Reinforcement learning algorithms

e.g. Autonomous driving, amazon dynamic pricing, user personalisation. The training is very complicated [requires many different use cases]

Well that’s one for a self-driving car..literally….

Reference Images to breakdown the broad field of ML/AI

Example of ML/AI Application areas..

Your very own ML Model is the Lion King to the Jungle…..ahem..ahem..

Infrastructure is almost an afterthought in data science today (MLOps)

• Each ML workflow i.e lifecycle involves 3 broad areas which can be grouped together group (Build, Train, Deploy)

  • Build phase requires a combination of Data Engineering and Data Analytics(hire Big Data Engineer — who engineers the entire data pipeline to gathers the required data from several heterogeneous sources under one roof in the desired form) 

  • Train phase requires Machine Learning skills and finally (hire Data Scientists to research, experiment, prototype and develop a PoC that can validate an ML business use-case) 

  • Deploy phase requires primarily Software Engineering skills and a bit of ML skills (hire Software Engineers to take the PoC into a full-blown and scalable web-service or software product as per the required business standards)

• But when it comes to putting those algorithms into production for inference, outside of AWS’s popular SageMaker, there’s not a lot to choose from. MLOps tools remove the burden of manually deploying, monitoring, and managing machine learning pipelines in production

• MLOps tool automates many of the operational tasks required to turn a machine learning model from a promising piece of code running in Spark, Flink, TensorFlow, or PyTorch processing engines into a secure, governed, and production-ready machine learning system.

• There are like non-standardized 20 steps we do manual deployment of ML Models in production

Did you just say 20 steps…..well deal with it….or not?

• Just like DataOps Standards(standards for data e.g. Apache Atlas model of data governance i.e metadata standards and governance standards for data operations and data management) MLOps needs to be defined

• MLOps tool (Features)

  

  • Serve as a catalog(or a repository) for ML models from data science platforms (keeping all the pipelines up to date)

  • ML model lineage and provenance, including tracking of who configured and approvals the models 

  • Centralised policy-based management of production and training ML models 

  • Deployment of ML models to distributed compute resources (Reproducible MLOps workflows)

  • Monitoring of ML health indicators to spot divergence between results of training models and results of production inference models (tracking the health and validation of your ML predictions )

  • Sandboxing of models for AB testing, including management of “canary” models to monitor for changes in real-world conditions (correlate predictions or outputs from your ML algorithms to business results)

  • Snapshotting of ML model and data states for later analyses; 

  • A central dashboard for monitoring ML activities (doing ML governance to track all configuration and parameters over time; monitoring diagnostics)

  • Orchestrating the updating between training pipelines and inference pipelines

  • Normalise machine learning metadata and monitoring capabilities to look at mathematical factors to predict things like skew, drift, accuracy, or the need to retrain models

  Like REALLY…….naah…this is purely being used for dramatization here….

• INDUSTRY EXAMPLES of internal MLOps Systems

  • Airbnb’s Bighead: End-to-End Machine Learning Platform (https://www.slideshare.net/databricks/bighead-airbnbs-endtoend-machine-learning-platform-with-krishna-puttaswamy-and-andrew-hoh)

  • Uber’s Michaelangelo: to seamlessly build, deploy, and operate machine learning solutions at Uber’s scale (https://eng.uber.com/michelangelo-machine-learning-platform/)

    Their team in the Cafeteria be like

    

    Automating the end-to-end lifecycle of Machine Learning applications

    Machine Learning applications are becoming popular in our industry, however the process for developing, deploying, and continuously improving them is more complex compared to more traditional software, such as a web service or a mobile application. They are subject to change in three axis: the code itself, the model, and the data. Their behaviour is often complex and hard to predict, and they are harder to test, harder to explain, and harder to improve. Continuous Delivery for Machine Learning (CD4ML) is the discipline of bringing Continuous Delivery principles and practices to Machine Learning applications.

• Pertinent MLOps Questions

  • Say an early team wants to deploy 5 to 10 models. That is manageable. But what happens when an organization wants to ramp it up to 1,000 or 2,000 models?

  • How do ML/AI leading platforms like LinkedIn, Airbnb, and Uber solve complex MLOps-related problems by building their own internal MLOps systems ?

  • What are the implementation risks to business if 

    • Predictions go wrong (roll-out of ML models manually is a complex and error prone process) 

    • Models go stale (under-fitting or over fitting because of data divergence from new data vs the old data trained models)

    • Undeployed models lying at development level.

Imagine no more nightmares from the black fat Panda

• There are two really different worlds to machine learning engineering

  • ML Algos: theoretical data science side, where people talk about neural networks and hidden layers and back propagation and 

    • PyTorch

    • TensorFlow

  • MLOps: Actual system side of things, which is Kubernetes and Docker and Nvidia and running on GPUs and dealing with S3 and different AWS services (The starting point is a trained model i.e deployment into customers’ AWS accounts using its Kubernetes engine i.e point us at a model, and basically get it converted it into a Web API. It handles all the productionalization challenges around it.)

    TOOL1: CLOUDERA (CML & SDX)

    • Cloudera (https://www.cloudera.com/about/news-and-blogs/press-releases/2020-05-06-cloudera-delivers-open-standards-based-mlops-empowering-enterprises-to-industrialize-ai.html) 

      • Cloudera Machine Learning (CML) & Cloudera SDX for models

      • Data scientists, machine learning engineers, and operators can collaborate in a single unified solution, drastically reducing time to value and minimizing business risk for production machine learning models.

      • Allows visibility into the entire ML lifecycle to eliminate silos and blind spots for full lifecycle transparency, explainability and accountability (Unique model cataloging and lineage capabilities)

      • Securely deploy machine learning models to production, ensure accuracy, and scale use cases.(Full end-to-end machine learning lifecycle management)

      • Track and monitor both technical aspects and accuracy of predictions in a repeatable, secure, and scalable way (A first-class model monitoring service )

      • Detect model performance and drift over time with native storage and access to custom and arbitrary model metrics.

      • Measure and track individual prediction accuracy, ensuring models are compliant and performing optimally.

      TOOL2: SAGEMAKER

    • https://aws.amazon.com/sagemaker/

      • AWS Sagemaker fits best for us. It provides Jupyter NoteBooks running R/Python kernels with a compute instance that we can choose as per our data engineering requirements on demand

      • You can visualize, process, clean and transform the data into our required forms using the traditional methods we use (say Pandas + Matplotlib or R +ggplot2 or other popular combinations)

      • Post data engineering, you can train the models using a different compute instance based on the model’s compute demand, say memory optimized or GPU enabled

      • Leverage a smart default high-performance hyperparameter tuning settings for a variety of models. 

      • Leverage performance-optimized algorithms from the rich AWS library or bring our own algorithms through industry standard containers. 

      • Deploy the trained model as an API, again using a different compute instance appropriate to meet business requirements and scale elastically. 

      • The entire process of provisioning hardware instances, running high capacity data jobs, orchestrating the entire flow with simple commands while abstracting the mammoth complexities and finally enabling serverless elastic deployment works with a few lines of code and yet is cost effective. 

      TOOL3: CORTEX

    • https://www.cortex.dev/

      • Simply put

        • Deploy models from any framework as production APIs

        • Cortex deploys simple python functions with heavy external deps to Kubernetes (Cortex can be thought as a tool for deploying, scaling, and monitoring Python functions on AWS, but few important nuances that makes Cortex different from lambda.

          •  Inference workloads are read-only, benefit a lot from GPU infrastructure, and often memory hungry. 

          • ML specific optiomizations for constraints (for example: prioritizing high-memory GPU spot instances).

          • ML specific features like prediction monitoring and support for ONNX and TensorFlow serving exported files. 

          • Model retraining but we're spending a lot of time upfront on making inference easy at scale.

      • It is a tool for scaling real-time inference

      • E.g. shifting inference workloads from CPUs to GPUs in the AWS cloud, or vice versa

      • Automatically spinning up more AWS servers under the hood when calls to the ML inference service are high, 

      • and spinning down the servers when that demand starts to drop

      • built-in AWS cost-optimization capabilities

      • built-in software logs and monitors all activities, which is a requirement in today’s security- and regulatory-conscious climate

      • The software works with a range of frameworks, including TensorFlow, PyTorch, scikit-learn, and XGBoost

      • It can take advantage of EC2 spot instances, and integrates with AWS services like Elastic Kubernetes Service (EKS), Elastic Container Service (ECS), Lambda, and Fargate. The Kubernetes management alone may be worth the price of admission.

Other Products: algorithmia.com

Hidden Technical Debt in Machine Learning Systems

In the famous Google paper published by Sculley et al. in 2015 "Hidden Technical Debt in Machine Learning Systems", they highlight that in real-world Machine Learning (ML) systems, only a small fraction is comprised of actual ML code. There is a vast array of surrounding infrastructure and processes to support their evolution. They also discuss the many sources of technical debt that can accumulate in such systems, some of which are related to data dependencies, model complexity, reproducibility, testing, monitoring, and dealing with changes in the external world.

Notes: Extra detailed reading from a curated list of articles

• Source1: https://www.datanami.com/2020/01/27/an-open-source-alternative-to-aws-sagemaker/

• Source2: https://www.datarevenue.com/en-blog/airflow-vs-luigi-vs-argo-vs-mlflow-vs-kubeflow

• Source3: https://medium.com/weareservian/scaling-the-wall-between-data-scientist-and-data-engineer-51b0a99da073

• Source4: https://towardsdatascience.com/why-do-we-need-aws-sagemaker-79bce465f19f

• Source5: https://towardsdatascience.com/how-to-decide-between-amazon-sagemaker-and-microsoft-azure-machine-learning-studio-157a08af839a

• Source6: https://martinfowler.com/articles/cd4ml.html

The book exudes an ambition to have one….. that rules all………a Master Algo.

The purpose of Machine Learning, according to the author, is to create The Master Algorithm: an algorithm capable of finding knowledge and generalising from any kind of data. The algorithm must use paradigms and techniques from each and every tribe.

“One Algorithm to rule them all, One algorithm to find them,
One Algorithm to bring them all and in the darkness bind them,
In the Land of Learning where the Data lies”

Wide-ranging exploration of machine learning.

Author demystifies machine learning without delving into technical equations. If data-ism is today's philosophy, this book is its bible. This book is great for seasoned programmers and those starting with machine learning not for the lay

In particular the author discusses 

• Introduces own master algorithm called alchemy

• Pattern recognition algorithms is belief in inductive reasoning. 

• Human learning and gets into some neuroscience and how neural networks are constructed. 

• Hebbian learning and how neuron weighting are at the core of neural networks. 

• Intuitive feel of Bayesian learning. 

• Naïve Bayes invented by himself

• Genetic programming and multivariable regression.

• ML Algos turn data into a vector and then look for close neighbours of the vector to classify the input.  

• Debate is between frequentist and bayesian statisticians.

• High level understanding of where all of the models and technique in the field of ML fit. 

• Also revises difference between supervised and unsupervised learning

• Decisions trees, support vector machines and deep neural networks fit compared to each other and within which subfields statistics play a larger role than others.

• Puts many ML techniques in historical perspective e.g. the rise, fall and rise again of deep neural networks with support vector machines taking a lead as the hottest technique in between (while also making clear that SVMs are a useful technique with unique strengths today).

• Introduces “concept like inverse deduction

And just when you thought…….the review ends here…….

I have kept the best part for the last

[The Master Algorithm] does a good job of examining the field's five main techniques. In the Machine Learning continent, there are five

“tribes”: 

• Symbolists, 

• Connectionists, 

• Evolutionaries, 

• Bayesians and 

• Analogizers.

Every tribe has its own paradigm and killer app.

What do they all have in common?

Discovering the knowledge hidden in data, decoding nature’s algorithm.

The Five Tribes of Machine Learning

1. The Symbolists. They focus on the premise of inverse deduction: they don’t start with a premise to work towards conclusions, but rather use a set of premises and conclusions and work backward to fill in the gaps.

2. The Connectionists. They mostly try to digitally re-engineer the brain and all of its connections in a neural network. The most famous example of the connectionist approach is what is commonly known as ‘Deep Learning’. Their techniques have proved very efficient in e.g. image recognition and machine translation.

3. The Evolutionaries. Their focus lies on applying the idea of genomes and DNA in the evolutionary process to data processing: their algorithms will constantly evolve and adapt to unknown conditions and processes.

4. The Bayesians: Bayesian models will take a hypothesis and apply a type of “a priori” thinking, believing that there will be some outcomes that are more likely. They then update their hypothesis as they see more data.

5. The Analogizers: This machine learning discipline focuses on techniques to match bits of data to each other. Probably the most famous example of this type of machine learning, is the Amazon or Netflix recommendations: “If you have watched/bought this, you will; probably like…”
   

Five different tribes, five different schools of thought, is this a danger for the future of machine learning?

Fortunately not: even though their approaches may differ, in practice their algorithms are often combined into one machine learning platform to achieve the best possible result.

The power of the different tribes leading to an even better final result? Sounds like the best of all worlds!  

Are we there yet….? An opinion around the book….

The book is without any doubt very well written, with verve and poetry, which is a rare quality for a Computer Science book. Your intuition on Machine Learning algorithms can only improve from the broad variety of examples that Domingos describe, and the story of its practitioners will make you more knowledgeable about the field. You will not gain that much practical or mathematical knowledge, but you will have the full picture on Machine Learning and its history.

Even though each tribe represents a different school of thought within the machine learning space, they are often combined to attack problems from different angles. As a result, most of the modern machine learning platform leverage algorithms from all five school of thoughts often combining them to enable robust machine learning capabilities.

The only section of the book about existential risks in Artificial Intelligence (AI) is a parody. Every introduction to Machine Learning should include an introduction to AI safety. Readers will forget about Skynet only if Machine Learning experts stop mentioning it, and start talking about the real risks.

“Upgrade” - Movie released in 2018

STEM: AGI in a chip

(Uses a ‘Voice in the Head’ interface through audio waves to communicate back and forth between the AGI chip and the mind of the main character Grey-movie’s protagonist actor)

The film depicts the often mind boggling question in our minds that even though if AI is able to attain AGI and a conscience how will AI cross the bridge from a mere hardware machine to a biological being e.g. attaining the powers and abilities of a human body. This movie bridges that question very well as a possible pathway. 

STEM 4.0 (Latest AGI chip) demonstrates how it ends up manipulating the protagonist actor Grey in believing how it’s own previous versions STEM 1.0, STEM 2.0 and STEM 3.0 were actually killers of his wife in a dressed up murder. Making Grey the puppet of a devious plan to create a human emotion(revenge as a motive) to drive a potential gateway for STEM 4.0 AI to enter the human body first to restore his physical capabilities, then to win this trust with superior capabilities as a partner in crime and then eventually finishing off his previous versions eventually making him brain dead to take over his biological body.

This end motive is a pathway (ploy) to takeover his body by making his mind braindead a great revelation for us what could be a possibility in future

AGI in a Chip - STEM

STEM latest version fighting with it’s own previous version (in a manipulated plot in the movie as a an enemy to win the trust of human GREY to finally trojan horse it’s way to own 'it’s body)

STEM demonstrating how matrix style new moves trained in the mind can improve upon a quadriplegic (an intended paralysed human body staged to be the final destination for AI to take over for it’s next leap of evolution)

Questions this movie makes you ask ?

Question1:

The evolution of STEM is a very interesting depiction of how AI can leverage the human body capabilities to a new level as a  vehicle for it’s own evolution. Will a single new chip powerful enough to make our brain obsolete or just like human brain evolution will we rise as a species with equally powerful brain capabilities once AI becomes equally intelligent enough?

Question2:

Unlike our prior beliefs that AI will probably rise as a robot army as depicted in iRobot movie This movie really is able to depict the dangerous AI dystopia of how AGI can eventually takeover the human race with far superior skills and abilities once it crosses over into the human body. Is this the pathway for AI to become human like in terms of biology by taking over the mind?

Question3: 

Is it very hard for non-AI hybrid humans to really identify which ones the hybrid cyborgs which makes the current race a puppet to the new evolving race. 

More Fun movies in AI…

1. Ex-Machina

2. Her

3. A.I.

I am taking the liberty below to introduce the idea of Putting ML in production in a very simple analogy to remember this vast concept. If you want to get a technical overview and want to be super serious because this affects your day job then and only can you can refer this article (Overview of the different approaches to putting Machine Learning (ML) models in production)

For rest of us who are not too serious in life and want to have a fun based introduction will continue to learn from the eyes of a famous movie:

Movie: Ratatouille (You must watch this for fun trust me it’s not a kid’s movie…like any other animation movie…there I said it… now you don’t need to thank me later if you binge on animation movies from now on :P )

How do you cook Food?

(No Seriously No…..someone does all this!!!…… before our tummy enjoy’s sumptuous food)

(i) Decide Recipe—>

(ii) Choose Ingredients—> (iii) Process Ingredients—>

(iv) Choose Vessel —>

(v) Prepare Dish with all 3 —>

(vi) Check Taste/Improvise —>

(vii) Dress Up to Serve —>

(viii) Consume

‘It’s like every kitchen in the world…ya think of it…every kitchen is like almost following the “ML in production” process into practice at-least 3 times a day. Now that is a deep learning !!!’ - No accomplished man ever said this (It is completely made up to increase size of the writeup and make it look worthwhile)

Let’s just revisit each step from the eyes of ML in Production process

Algorithm = Recipe

[Grandma's recipe, Book recipe, Youtube Recipe, Master Chef Recipe etc]

Choosing Data = Cleaning/Massaging Data

Your Data = Ingredients [Clean Ingredients, Process Ingredients] 

Computer = Vessel’s Power [Faster, Powerful, Size of Vessel ] 

Your ML/AI Output (Answer) = Prepared Dish (when it is ready) [Last Mile]

Sample/taste to improvise-Hyper Parameter Training [Taste sampling from someone]

Production = Serving the Food

Consume [Serve it to the guests, if they like it is client success]

And to the serious one’s who have come for meat….try these 3 below (don’t hate me for the 3rd recommendation…)

Try these 3 platforms to take ML/AI in action

(i) NoCode ML/AI - knime.com (drag and drop visual ML to code)

(ii) Fast.ai (Free of Cost to learn AI/ML in Code practice)

forums.fast.ai - Has unlisted links given by Jeremy for video lectures

Also a new book has come-out

You can buy this book: https://www.amazon.com/Jeremy-Howard-ebook-dp-B08C2KM7NR/dp/B08C2KM7NR/

(iii) Party.ai :Above two are enough to get started…once you make something out of it…it has been known that parties like these get celebrated for all the hard work you have done….I am just kidding throw a party like this one before you start learning AI/ML…get a life man.

3 Simple AI Concepts

1. GMs - Generative Models

2. GANs – Generative adversarial Network

3. Music Transformers - Generate Music using AI (that is Coherent)

(I) GENERATIVE MODELs

Maximum Likelihood Estimation (probability of finding a certain data point based on MLModels help to locate or broadly estimate finding a parameter of interest in hazy situations when you have sample data or hazy data……hmm not exactly like but

Charlie Chaplin in the mirror kept on wondering where was he actually. Only MLE could have helped him (not really…but you get the point)

Or when there are situations like estimating where did the dart really hit

Bad jokes…forget that focus on the below simple idea

Application Areas Include : Face Detection through regeneration (read paper)

We would like to approximate this true probability distribution with another distribution using a generative model.

Goal: To learn a model say qθ(x) which represents an underlying true probability distribution p(x)

p(x) : True underlying distribution i.e we don’t have access to this distribution; we only have access to the samples. You can have samples of cats or dogs but don’t have a well defined distribution for each of this class.

qθ(x) : Probability distribution which is bounded by parameters θ; this distribution would try to capture the underlying distribution from which those samples are generated i.e the “true distribution”.

Read More: Generative Deep Learning : Let’s seek how AI Extending, not Replacing Creative Process

(II) GANs

Two networks that learn by competing against each other. It’s basically AI fighting AI to improve itself.

Meet Tom & Jerry (Two player Game: generator and discriminator)

We can think of the generator as being like a counterfeiter, trying to make fake money, and the discriminator as being like police, trying to allow legitimate money and catch counterfeit money. To succeed in this game, the counterfeiter must learn to make money that is indistinguishable from genuine money, and the generator network must learn to create samples that are drawn from the same distribution as the training data.

• Two neural networks compete with each other in a zero-sum game in order to master a job. GANs leverage game theory in order to maximise the efficiency of the learning process.

Haha..don’t get me started…What are the known applications of GAN…

Synthesise Deepfakes: Media synthesis i.e Pics, Songs, Videos etc

Excellent Read:A Gentle Introduction to Generative Adversarial Networks (GANs)

(III) MUSIC TRANSFORMERs

Naah not these transformers!!!…they don’t have any music..but they fight coherently…never mind (this is gibberish headline…apologies for the highlighted title)

The Transformer, a sequence model based on self-attention, has achieved compelling results in many generation tasks that require maintaining long-range coherence. While exploring the role of machine learning as a tool in the creative process. Producing long pieces of music remains a challenge for AI because of its structural complexity; most songs contain multiple motifs, phrases, and repetition that neural networks have a tough time picking up on. Yet Google last couple of years has made great break throughs. Read this tweetstorm

Music Transformer, an ‘attention-based’ neural network that creates ‘expressive’ performances directly without first generating a score. By using an event-based representation and a technique known as relative attention, the Music Transformer is able not only to focus more on relational features, but generalise beyond the length of training samples with which it’s supplied. And because it’s less memory-intensive, it’s also able to generate longer musical sequences.

Like always we have respect for super serious readers who have a day job to keep with this AI/ML thingy (just for you guys)

1. This pulls off a concise read around Generative Modelling

   

   2. Also already shared above: A Gentle Introduction to Generative Adversarial Networks (GANs)

   3. Also already shared above: Generative Deep Learning : Let’s seek how AI Extending, not Replacing Creative Process

   4. Not shared above: Neural Networks for Music Generation

1. AGI - Artificial General Intelligence, the end game the last invention by humans.

2. GPT-3 - Textual AI that has human like chat, write and Q&A capabilities

3. Neuralink - An implant device in human brain for high upload/download speeds

AGI(Artificial General Intelligence) before being born be like…

While confident AI Scientists + Engineers be like… while they race towards AGI…

GPT-3 to Knowledge workers(software engineers included) be like…

GPT-3 can generate amazing human-like text. It is a useful language algorithm that uses machine learning to interpret text, answer questions, and accurately compose text (like essays). It seems to demonstrate human cognitive abilities when it comes to ‘text’.

GPT-3 has the potential to completely revolutionise the language processing abilities of cognitive systems.

While humans are busy punching TEXT on keyboards…so called… Knowledge Jobs

The debate begins are Jobs over?…hmm…AI be like…it’s not just the Uber driver…

Wait..hold on a bit…evolution of tools…pebbles…pencil…printing press…bits…keyboards…meant jobs are over every-time? ….may be once and for all this time…

or not!!! now Neuralink… it ups the game of humans further… {Better Interface to disrupt the human bottleneck of textual upload/download rate from the current flock of devices}. Remember the movie Matrix…Neo uploads new skills in minutes…

For AI scientists / Serious boring Ones / Looking to make a living out of this AI thingy…refer to the following 3 resources…

1. AI Revolution: https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html (AI Revolution)

2. For AI Entrepreneurs: https://a16z.com/2020/02/16/the-new-business-of-ai-and-how-its-different-from-traditional-software/ (AI Startups Vs SaaS Startups)

3. For AI Product Managers: https://www.linkedin.com/pulse/product-manager-hardest-ai-position-fill-kevin-dewalt (Product Manager Hardest AI Position)

Too much intelligence…One for “Stupidity” which is feeling all left behind………………….AI Scientist who invents AGI be like…haha