Quick answer. Two of the nine RBI Grade B DSIM syllabus units are Data Science, Artificial Intelligence and Machine Learning and Database and Data Warehouse Management. Together they name recurrent and convolutional neural networks, natural language processing, gradient boosting, support vector machines, SQL joins, star and snowflake schemas, ETL and OLAP versus OLTP. Both Paper I and Paper II draw from all nine units, so neither can be skipped. Most Statistics postgraduates have studied none of it.
Read the Syllabus Line, Not the Cadre Name
DSIM stands for the Department of Statistics and Information Management. The name suggests a classical Statistics examination.
Open the official syllabus and units 8 and 9 read like a machine learning course catalogue.
Unit 8, Data Science, Artificial Intelligence and Machine Learning Techniques:
Supervised and unsupervised pattern classification. Reinforcement learning. Model accuracy measures. Linear regression, logistic regression, penalized regression, naive Bayes, nearest neighbour, decision tree, support vector machine. Kernel density estimation and kernel discriminant analysis. Neural networks. Kernel regression, tree and random forests. Hierarchical and non hierarchical clustering with k means, k medoids and linkage methods. Cluster validation indices including the Dunn index and Gap statistics. Bagging with random forest and boosting with adaptive and gradient boosting. Recurrent Neural Network. Convolutional Neural Network. Natural Language Processing. Recursive feature elimination, variance inflation factor, ensembling and stacking, Elastic Net regularization, hyperparameter tuning via grid search, feature importance interpretation and cross validation strategies.
Unit 9, Database and Data Warehouse Management:
Data structures. RDBMS and NoSQL fundamentals. Database normalization, redundancy elimination, consistency. SQL for querying, updating, aggregating and managing relational data. Database joins: inner, left, right and outer, with applications to data merging. NoSQL overview: document based, key value, wide column and graph databases. Data warehousing, star and snowflake schemas, ETL processes, OLAP versus OLTP. Database indexing and optimization. Basics of big data frameworks.
That is a working data scientist’s toolkit, examined by a central bank.
Why RBI Did This
The official job description explains it, and it is worth reading because the interview tests whether you understand the role.
The Board says the DSIM role involves collecting, compiling, analysing and interpreting data on the banking, corporate and external sectors, with heavy use of advanced statistical and econometric techniques to model and forecast inflation and growth. It names classical and Bayesian inference, high dimensional regression, time series models and simulation based methods.
Then it goes further. Candidates are expected to manage large scale structured and unstructured datasets using various databases, and to develop, maintain and optimise queries for data integration and reporting. The role calls for applying data science, artificial intelligence and machine learning techniques to extract insights from voluminous datasets.
The syllabus follows the job. A central bank that forecasts inflation from large administrative datasets needs officers who can write a query, build a model and explain the result. Units 8 and 9 exist because the work exists.
Eligibility Confirms It
You can now qualify for DSIM with a Master’s degree in Data Science, Artificial Intelligence, Machine Learning or Big Data Analytics, with 55 percent, as an alternative to a Statistics degree.
That is not a footnote. It is the Reserve Bank saying a machine learning postgraduate is as welcome in its Statistics cadre as a Statistics postgraduate.
Which means your competition has changed. Some of the people writing this paper studied neural networks formally for two years. If you are treating units 8 and 9 as optional, you are competing against people for whom they are the easy half.
What This Means for Your Preparation
Both papers draw from all nine units. Paper I is objective and Paper II is descriptive, but the pool is the same. There is no unit you can quietly avoid.
Nine units and ten vacancies. A candidate who can answer seven units well and two not at all is at a real disadvantage against one who can answer all nine adequately.
The honest scale of the work
| Unit | If you are a Statistics postgraduate |
|---|---|
| 1 to 5 | Revision, with some new named topics |
| 6 | Partly new. ARCH and GARCH are usually beyond a Master’s syllabus. |
| 7 | The optimization half is familiar. MCMC, Gibbs sampling and Metropolis Hastings usually are not. |
| 8 | Mostly new. Six to eight weeks. |
| 9 | Almost entirely new. Two to three weeks. |
Unit 9 is the shorter of the two and it is largely definitional. SQL joins are learned by writing queries for a weekend. Star versus snowflake, ETL and OLAP versus OLTP fit on one page of notes.
Unit 8 is the real work, and it does not compress.
Where to Start
Start with Unit 9, not Unit 8. It is smaller, it is finishable in two weeks, and finishing something builds the momentum you will need for Unit 8.
Then Unit 8, using An Introduction to Statistical Learning, which is free to download legally from its authors and is written for readers with a statistics background. It covers most of what the syllabus names.
For recurrent networks, convolutional networks and natural language processing, use short focused reading rather than a full course. The syllabus asks for concepts, not implementation. You will not be asked to train a model. You may well be asked what distinguishes a recurrent network from a convolutional one.
Read the full nine unit syllabus
The Mistake to Avoid
The instinct is to master the familiar units first and leave the unfamiliar ones for later.
That instinct is why candidates arrive at the examination able to answer six units out of nine. Familiar material compresses under time pressure. New material does not.
Put units 8 and 9 second in your plan, not last.
Frequently Asked Questions
Is machine learning really in the RBI DSIM syllabus?
Yes. Unit 8 explicitly names logistic and penalized regression, naive Bayes, nearest neighbour, decision trees, support vector machines, neural networks, random forests, adaptive and gradient boosting, recurrent and convolutional neural networks, and natural language processing.
Is SQL in the RBI DSIM syllabus?
Yes. Unit 9 covers SQL querying, updating and aggregating, all four join types, NoSQL databases, star and snowflake schemas, ETL processes and OLAP versus OLTP.
Can I skip the data science units in RBI DSIM?
Both Paper I and Paper II draw from all nine units, so skipping two of nine leaves a significant part of the paper unanswerable.
Do I need to write code for RBI DSIM?
The papers are objective and descriptive, not practical. The syllabus tests conceptual understanding of the techniques rather than implementation.
Can a Data Science postgraduate apply for RBI DSIM?
Yes. A Master’s degree with 55 percent in Data Science, Artificial Intelligence, Machine Learning or Big Data Analytics is one of the three eligibility routes.
Which book covers the machine learning unit?
An Introduction to Statistical Learning by James, Witten, Hastie and Tibshirani, which is free to download from the authors and is written for readers with a statistics background.
