that I have stumbled across while trying to learn about AI and ML. I put it together in a manner that makes it easy for someone new to this field to learn and start thinking about it from a product stand point
Predictive Content Recommendation CRM and Marketing Automation Customer Service and Customer Success E-commerce Core Value Proposition shift to AI and ML Building Smart customer success platform Intelligent Shopping Platform Amazon Go: A store of the future Personalized cheap travel deals Automatic answers to power customer service Aiding smart conversations
characteristic of human intelligence “- John McCarthy, 1956 “ Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed” by observing data -Arthur Samuel, 1959 Deep learning is one of many approaches to machine learning. Deep learning was inspired by the structure and function of the brain, namely the interconnection of many neurons.
and produce outputs. They require labeled data upfront ! Category Problem Type Example Regression Point in time prediction Lead conversion, Price Optimization Forecasting Sales forecasting Classification Binary Classification Will this post go viral ? yes/no Multi-label classification Document tagging, photo tagging
No need to tag the data set. Problem Type Example Clustering Identify similarity criteria and find that data that match it. Eg. Twitter Trends Association People who brought X also bought Y Anomaly Detection Fraud Detection
need some training data sets but not as much as supervised learning - Are used to solve the same type of problems as supervised and unsupervised learning This is how human babies learn !
data and learns as it gets more feedback about its predictions over time. - The correct input/output is never presented nor sub-optimal predictions are explicitly corrected. - Involves defining the concept of reward Focus is to strike balance between exploration (uncharted territory) and exploitation (of current learning)
to get better results than an individual model ➔ Natural Language Processing ◆ Field of computer science dealing with understanding language ◆ Types of problems solved combining NLP and ML • Keyword generation • Language Disambiguation • Sentiment Analysis • Named entity extraction ◆ Used in chatbot applications and preparing datasets
customization of the system and user experience - Optimized price - Personalization - Recommendation Automate Tasks - Data entry into CRM system. - Filing expense reports Automatic retrieval, generation or processing of content - Ranking and relevance of search results. - Finding relevant conversations Predictions, estimates and trends at scale - Sales Forecasting - Identifying trends in renewal threats - Problems in sales pipelines - Customer Segmentation Detection of unusual activity or system failure - 1:1 reporting on unusual system usage - Resource utilization issues in cloud environment
for your customers ◦ Do I have clear understanding of my customer segments ? ◦ How can I identify good customer experience from bad customer experience ? ◦ How can I personalize my end user’s experience with the product ? ◦ What are the decisions and choices I’m asking my customer to make today ? Can those choices be automated ? • Improvements in internal functions, processes and business logic ◦ What type of data do people in my own company work with ? Can we automate gathering and working with this data ? • Expansion to new verticals and new products ◦ Can we create brand new products for existing customers ? ◦ Who else can benefit from our data and insights ?
small change in the problem definition could mean an entirely different algorithm or model. - Keep explainability and interpretability in mind while defining use cases - Defining the behavior of the product around ML is just as important as the accuracy of ML models
role in creating ML models - Steps involved in modeling - Ideation - Align on the problem - Define potential metrics - Identify potential inputs - Data preparation - Identify ways to collect data - Data cleanup, sanitization, anonymization - Prototyping and testing - Identify training and validation data sets - Identify when the model fails - Productizing - Collecting data at scale - Data governance
Generalization : How well the model can be applied to specific examples not seen by the model while it was learning - Overfitting : When the model learns the model too well that it negatively affects model’s performance - Precisions : % of positive predictions that are correct. The higher the precision the less the false positives - Recall: % of true positive in the data that was identified by the algorithm as positive. The higher the recall the lesser the false negatives. - If you want higher precision, your recall will be lower. The model will “prefer” to label something as a negative than label it wrongly as positive.
about winning the trust of your end user. Hence account for fallback or non-ML scenarios. - Test your model as much using backdating. - Consider exposing some of the underlying data as part of your user experience design. - Show only results that facilitate easy decision making and not hundreds of stack ranked results (classic example: Amazon’s product recommendation that shows three choices i.e. best suited, top rated, cheapest)
generic. It includes things like planning, understanding language, recognizing objects and sounds, learning and solving problems. • Can be classified into two categories ◦ General AI : Emulating all characteristics of human intelligence ◦ Narrow AI : Emulating some facets of human intelligence • AI can be achieved without ML but that would involve millions of lines of code with complex rules and decision-trees.
the ML approaches. Other approaches include decision tree learning, inductive logic programming, clustering, reinforcement learning , Bayesian networks and many more • Artificial Neural Networks (ANNs) are algorithms that mimic the biological structure of the brain. • In ANNs there are ◦ Neurons, which have discrete layers. Each layer picks out a specific feature to learn. ◦ Connections to neurons • It is this layering that gives deep learning its name.