VIEW POINT

LINEAR ALGEBRA IN AI

Abstract
As the BFSI industry begins to realize the benefits of artificial
intelligence, this whitepaper introduces linear algebra as a
computational tool and its advantages when compared to conventional
artificial intelligence technology. Linear algebra belongs to a branch
of mathematics that improves machine learning algorithms and data
analytics. This paper also presents a couple of potential use cases
and represents how linear algebra will be the potential growth path,
foreseeing the increasing acceptance of AI-enabled solutions.
Why linear algebra?

Linear algebra plays a vital role in better tensors, and scalars. This paper will explain learning and artificial intelligence
understanding artificial intelligence and these in detail. (deep learning) by exposing itself to
quantum computing, as currently BFSI Data is an important factor in the outcome data clustering, data classification, data
is heading towards advanced tools and of any event. The four pillars mentioned validation, and data fitting. Any time-
techniques of deep learning, a subset of above will be used as repositories to store consuming deep analysis of data requires
artificial intelligence. Linear algebra, widely data and operate on it as per provided storage, and linear algebra provides the
considered as the “guru of mathematics”, instructions. Linear algebra by large is perfect solution for this, by using vectors,
is a computational tool for science, considered as a “storage space”. matrices, tensors, and scalars as a central
engineering, and data analytics. Four pillars data repository.
In simple terms linear algebra can be
of linear algebra are vectors, matrices,
widely used as a “data guru” for machine

The history of linear algebra

The year 1637 is when linear algebra geometry. Use of determinants to solve dynamic values in a data set has been
was first introduced. The concept of the linear systems was found in 1693 by a discovered by Irish scientist William Rowan
coordinate system was developed by German mathematician. This was further Hamilton in 1843.
Rene Descartes, who founded a geometric enhanced as Cramer’s Rule in the year
approach known today as Cartesian 1750. However, the complex theory of

External Document © 2023 Infosys Limited

Mathematical objects in linear algebra

Scalar: This is a real or natural number. Matrix: A matrix is used to identify each carried across all the other objects, such as
element in a number of arrays in two scalars, vectors, and matrices. An array of

[ ]4 dimensions; each element in a matrix is

ascertained by two indices. An example is
elements are arranged on a variable axis.
For instance, in a typical software coding
Vector: This represents an axis that requires
given below. The matrix is a combination language, a tensor is used as a variable to
a coordinate for each element. To identify
of two indices; one is for rows, which are 48 declare multiple data sets and extract the
each element, focus on an index where a
and 36, and the other is for columns, which information from a given set of files.
list of numbers is arranged in an order. A
are 43 and 86
simple vector is used to identify points in a
given space. This is a typical mathematical
object which is widely used in current [ ]48
36
machine learning techniques. Tensors: Unlike a matrix which is a

[ ]4 two-dimensional array, a tensor is an

3 n-dimensional array. The generic nature is

Linear algebra: A bird’s-eye view

Learning via ML Knowledge and

(Machine Learning) Reasoning

Problem Solving Planning Communication and NLP

Stored Data Points From Multiple Sources

Matrix Scaler

Vector Computational Tool Tensor

Data Source Database Database Data Source

Linear Algebra

Fig 1: The linear algebra computational tool

External Document © 2023 Infosys Limited

In the current BFSI industry we deal with Cloud Vision API, and IBM Watson Visual On the other hand, we also have a few top
high intense documents, which are a Recognition. These tools at the current data visualization tools in the BFSI industry,
combination of structured and dynamic business requirements have fairly matured such as Tableau, Google Chart, Jupyter,
documents, to validate customer details to perform auto identification of subjects and Power BI. These tools, in terms of the
in a typical onboarding process. The and objects in an image and unify the current business requirements, are fairly
document may include pure dynamic data images in a logical list. Having said that, on matured and provide a range of advanced
received from a customer via multiple these tools we still see some disadvantages databases as well as data visualizations
channels. which are mentioned below. with the help of big data applications,
We all know that there are several tools and 1) Misidentification of images artificial intelligence, and machine
techniques which are already introduced learning.
2) Lack of image clarity and unclear
in the markets for image recognition and background To further enhance the capabilities of
data visualization. Firstly, let’s understand these tools, we can take advantage of
3) Data insecurity or exposures
a few top image recognition tools in the linear algebra with its most common five
4) Technological imperfections
BFSI industry, such as Azure Content advantages mentioned below.
Moderator, Amazon Rekognition, Google 5) Increase In fraud

Data Set

Images and Photographs

Advantages Recommended Systems

Deep Learning

Later Semantics Analysis

Fig 2: Advantages of Linear Algebra

1) Data and its importance photographs. In BFSI, digital onboarding they occur. This output helps the machine
In any organization, data is very important. is the best example where images and learning model to easily compare queries
Linear algebra utilizes its capabilities in photographs of customers are validated. and produce appropriate responses.
machine learning as a model using data Machine learning applies linear algebraic 4) Recommended systems
sets with a tabular structure of rows and notations and operations to perform
Machine learning applies predictive
columns. In a typical BFSI example where additional actions such resizing, scaling,
analysis of linear algebra. This is currently
“free format” messages are received for and cropping for the desired result.
available with a fair amount of output. As
corporate actions or events, a matrix data 3) Latent semantic analysis an example: We see “Recommendations for
set can be applied. In this case, the free When words and text are involved in you” in major apps such as Amazon Prime
format message is considered as input in a any domain then the only solution for and Netflix.
matrix to identify observations, and after transforming the sentences is natural 5) Deep learning
analysis the output will be stored in various language processing (NLP). A contact
As mentioned earlier tensors are multi-
formats (important texts extracted from center is a typical BFSI example where
dimensional matrices that is used in
free format messages). customers’ dynamic requests are received
deep learning artificial neural networks.
2) Images and photographs via multiple channels. NLP large metrices
ANNs utilize larger datasets to work on
Linear algebra applies a table structure are used to store words, paragraphs, pages,
challenging tasks such speech recognition,
view in machine learning which will further etc., in a matrix and marks total words in
machine translation, etc.
apply the matrix algorithm on images or the form of a count or frequency in which

External Document © 2023 Infosys Limited

Use case 1: Handling text data using LSI (latent semantic indexing)

Typically, in BFSI operations, resources dynamically written text, generated as part flow diagram represents the current
must read and understand free formats of a case workflow. This is time-consuming transformation of these texts.
received as part of corporate events and and requires manual effort. The below

Free format Automation tool

message generated extract the raw
data

Corporate Action

Onboarded CA Event? Data extracted to (Applies latent Semantic Indexing)

successfully input file

Customer/Client Natural Language

Processing (ML)
Raise dynamic text Data extracted to
request in desired output End
workflow
Linear
Algebra

Automation tool Data extracted to

extract the raw input file
data
• Single Value Decomposition
• Text Document Matrix
• Identify Additional Stop Words
• Applies Multi Dateset Layers
Fig 3: Text data (latent semantic indexing)

With linear algebra in collaboration with be much more effective and can increase decomposition” and “text document
NLP (natural language processing) the data accuracy. The main components that matrix”, which are multi-dimensional
output from the desired input format will provide these results are “single value tensors.

Use case 2: Digital onboarding (image recognition and matching)

In BFSI operations KYC is the key process digital onboarding. Although there are flavor to mark up the received images with
for end-to-end onboarding of a customer. numerous tools in the market to perform additional information. The below picture
Image validation is a key element in image validation, linear algebra has its own depicts the importance of it.

Upload documents Image to image Embedded Machine (Applies latent Tri Dimensional Matrix)
for KYC process validation Learning (Python)
code

Customer/Client Applied Python Code

(ML)

• Convolution technique
• RGB Matrix
Linear
• Color extraction via channel Algebra
• Cropping, scaling and resizing

Fig 4: Tri-dimensional image recognition

External Document © 2023 Infosys Limited

With the application of linear algebra, we each element from pixel to local space and via different channels. This helps recognize
gain some additional advantages, such vector (multidimensional matrix) to RGB the image and process it as required
as convolution technique being added to conversion, which helps in color extraction through cropping, scaling, and resizing.

Going forward: AI hand in hand with linear algebra

Not all conventional approaches of artificial high complexity. Still, one must truly push think of Plan A and Plan B; however, with
intelligence provides desired results in towards linear algebra as a primary tool, linear algebra, we can derive executable
the context of BFSI, keeping in view of the considering its depth analysis with the help Plans A, B, and C. It can surely enhance the
dynamic nature of the processes and the of notations, arithmetic, various matrices, capabilities of modern machine learning
Documents associated with them. There statistics for data representation, and most techniques and also speech to text
are constraints in linear algebra as it is importantly the multi-dimensional matrix implementations such NLP, NLU and NLG.
a large mathematical component with approach. While building a solution, we

External Document © 2023 Infosys Limited

 Authors
Y Karteek Sarat Chandra
Senior Consultant, Infosys BPM

Karteek is a Senior Consultant for Digital Transformation Services at Infosys BPM. His core areas of
expertise are Capital Markets (Foreign exchange) Wealth Management, Private Equity, Reconciliation,
Stocks and Bonds. His primary focus is on designing digital solutions for the Capital Markets and identify
the automation levers to further resolve client challenges/ problems. His key areas of competence is in
Capital Markets Industry . He has previous experience in Project Management, Consultant and Business
analyst for RPA and analytics implementation.

Sourav Ghosh
Senior Industry Principal, Infosys BPM

Sourav is a Senior Principal with Infosys BPM’s Digital Transformation Services, responsible for Financial
Services & Insurance — Digital solution design and service delivery. An IBM-certified design thinking
practitioner, he advises organizations on their operations strategy, assists them in improving profitability
and efficiency of business processes, and helps in executing business transformation through calibration
of operating model and technology.

*For organisations on the digital transformation journey, agility is key in responding to a rapidly changing technology and business landscape. Now more
than ever, it is crucial to deliver and exceed organisational expectations with a robust digital mindset backed by innovation. Enabling businesses to sense,
learn, respond, and evolve like living organisms will be imperative for business excellence. A comprehensive yet modular suite of services is doing precisely
that. Equipping organisations with intuitive decision-making automatically at scale, actionable insights based on real-time solutions, anytime/anywhere
experience, and in-depth data visibility across functions leading to hyper-productivity, Live Enterprise is building connected organisations that are innovating
collaboratively for the future.

For more information, contact [email protected]

© 2023 Infosys Limited, Bengaluru, India. All Rights Reserved. Infosys believes the information in this document is accurate as of its publication date; such information is subject to change without notice. Infosys
acknowledges the proprietary rights of other companies to the trademarks, product names and such other intellectual property rights mentioned in this document. Except as expressly permitted, neither this
documentation nor any part of it may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, printing, photocopying, recording or otherwise, without the prior
permission of Infosys Limited and/ or any named intellectual property rights holders under this document.

Infosysbpm.com Stay Connected