Puducherry Technological University

Zeroth-Review

Image-to-LaTeX Mathematical Expression Solver:

Presented By
Under the guidance of,
Kondi Nanda Gopal Umesh Raju
Dr. Sagayaraj Francis,
Vengadesa Boopathi P
Professor
Gopesh
 Introduction

Image-to-LaTeX Mathematical Expression Solver:

• 1. The project presents an innovative approach to solving mathematical expressions by

combining image processing, LaTeX formatting, and user input, eliminating the need for manual
equation typing.

• 2. The system employs optical character recognition (OCR) technology to extract mathematical
expressions from images, converting them into standardized LaTeX format for accurate
representation of mathematical notation.
• 3. By using LaTeX as an intermediate step, the system ensures precise interpretation of
mathematical symbols and structures before proceeding to computation, reducing errors in
expression recognition.

• 4. Users can input specific values for variables identified in the expression, allowing for dynamic
solving of algebraic equations while maintaining the original expression's structure.

• 5. The solution bridges the gap between physical and digital mathematical processing, making it
particularly valuable in educational and research settings where quick verification of
mathematical expressions is needed.
 Literature Survey

Title, Year, Published in Methodology Merits Demerits

Title: "Image-to-LaTeX Trained a vision encoder- Demonstrated effectiveness The model's performance on
Converter for Mathematical decoder model using a Swin in converting both machine- complex, densely formatted
Formulas and Text" Transformer encoder and a generated and handwritten equations was not
Year: 2024 GPT-2 decoder to generate formulas into LaTeX code. extensively evaluated
Published in: arXiv LaTeX code from images of
mathematical formulas and
text.

Title: "Advancing OCR Assessed the application of Improved OCR accuracy by Focused primarily on OCR
Accuracy in Image-to-LaTeX active learning strategies to integrating active learning accuracy; did not extensively
Conversion" enhance natural language and LaTeX syntax address the subsequent
Year: 2023 processing-based OCR constraints. solving of extracted
Published in: MDPI Applied models for image-to-LaTeX expressions.
Sciences conversion.
 Literature Survey

Title, Year, Published in Methodology Merits Demerits

Title: "ConvMath: A Introduced a convolutional Demonstrated high accuracy The approach's applicability
Convolutional Sequence sequence modeling network and efficiency in recognizing to handwritten expressions
Network for Mathematical combining an image encoder printed mathematical was not addressed.
Expression Recognition" for feature extraction and a expressions.
Year: 2020 convolutional decoder for
Published in: arXiv sequence generation.

Title: "EDSL: An Encoder- Developed an encoder- Achieved high accuracy in The model's performance on
Decoder Architecture with decoder model with symbol- recognizing printed handwritten or complex
Symbol-Level Features for level features, including a mathematical expressions. layouts was not explored.
Printed Mathematical segmentation module and a
Expression Recognition" reconstruction module with a
Year: 2020 position correction attention
Published in: arXiv mechanism.
 Literature Survey

Title, Year, Published in Methodology Merits Demerits

Title: "Translating Math Proposed a deep neural Achieved state-of-the-art The model's effectiveness on
Formula Images to LaTeX network model with an performance on the handwritten or noisy images
Sequences Using Deep encoder-decoder IM2LATEX-100K dataset. was not thoroughly
Neural Networks" architecture, using a CNN evaluated.
Year: 2019 encoder and a bidirectional
Published in: arXiv LSTM decoder with
attention mechanisms.

Title: Deep Learning for Developed a transformer- Outperformed Mathematica Solutions require external
Symbolic Mathematics based seq2seq model for and Matlab in accuracy. verification (e.g., SymPy).
Year: 2019 symbolic integration and Generalized well to Struggles with out-of-
Published in: arXiv differential equation solving, symbolic expression distribution equations.
using large-scale manipulation. Computationally expensive
synthetically generated Leveraged deep learning for to train.
datasets. tasks traditionally handled
by rule-based systems.
 Literature Survey
Title, Year, Published in Methodology Merits Demerits

Title: An Integrated Platform Developed a LaTeX-based Simplifies model Limited to LP, MILP, and
for Intuitive Mathematical platform that converts development and debugging. MIQP models.
Programming Modeling Using mathematical models into Cross-platform, non- Requires basic AML
LaTeX Pyomo AML, integrating a
commercial, and reduces knowledge.
Year: 2018 Python parser and solver support
Published in: PeerJ via a user-friendly GUI programming effort. Lacks support for nonlinear
Computer Science Enhances reproducibility in and differential equations.
scientific research.
 Existing Work
Im2Text for Mathematical Expression Recognition (2017)
Research project by Harvard University
Uses deep learning for mathematical expression recognition
Converts images to LaTeX format
Does not include solving capabilities

MathPix (2016-Present)
Recognizes mathematical expressions from images and converts them to LaTeX
Focuses primarily on LaTeX conversion rather than solving
Offers API services for integration
Used widely in academic and research settings

Microsoft Math Solver (2019-Present)

Uses image recognition to solve mathematical problems
Provides multiple solution methods
Includes graphing capabilities
Does not output LaTeX format
Churn Prediction Flow Diagram
 Objectives
• Enhance Prediction Accuracy:
Use advanced ensemble techniques to combine multiple models, improving the accuracy of
churn predictions and helping businesses better identify at-risk customers.
• Increase Model Transparency:
Implement explainable AI methods like SHAP to provide clear insights into why customers
may churn, enabling businesses to understand the key factors driving the decision.
• Develop Targeted Retention Plans:
Use the improved predictions and insights to create personalized retention strategies, focusing
on customers who are most likely to leave, and take actions to retain them.
 Proposed Work

• Stacking and Boosting Models for Better Accuracy:

Using a mix of different models makes predictions more accurate. This method helps
combine the strengths of each model for better results.
• Using SHAP for Easy Understanding:
SHAP helps explain how each factor affects the churn prediction. It makes it easier for
businesses to understand why customers may leave.
• Better Retention Strategies with Advanced Methods:
Combining these techniques leads to more accurate predictions and clear explanations.
Businesses can then focus on customers who are most likely to leave and keep
them.
Churn Prediction Stacking
Churn Prediction Boosting
Churn Prediction SHAP
Working of SHAP
 Conclusion

• Our work aims to make churn predictions more accurate using advanced techniques like ensemble
models and explainable AI.
• This will help businesses understand why customers might leave.
• By identifying at-risk customers, companies can create better retention plans.
• Ultimately, this approach helps businesses keep more customers and improve their results.
 Reference
1. A Review on Machine Learning Methods for Customer Churn Prediction and Recommendations for
Business Practitioners
Year: 2024 | Journal Paper | Publisher: IEEE
2. Customer Churn Prediction System: A Machine Learning Approach
Year: 2023 | Journal Paper | Publisher: Springer
3. Analysis and Prediction of Churn and Retention Rate of Customers in Telecom Industry Using Logistic
Regression
Year: 2024 | Conference Paper | Publisher: IEEE
4. Customer Churn Prediction based on Stacking Model
Year: 2023 | Conference Paper | Publisher: IEEE
5. Research on a Customer Churn Combination Prediction Model Based on Decision Tree and Neural
Network
Year: 2020 | Conference Paper | Publisher: IEEE
 Reference

6. Customer Churn Prediction Based on Interpretable Machine Learning Algorithms in Telecom Industry
Year: 2023 | Conference Paper | Publisher: IEEE

7. Analysis of Customer Churn Prediction in Telecom Industry Using Decision Trees and Logistic
Regression
Year: 2016 | Conference Paper | Publisher: IEEE

8. Telecom Sector Churn Prediction Using Decision Tree and Random Forest Models
Year: 2023 | Conference Paper | Publisher: IEEE

9. A Comparison of Machine Learning Algorithms for Customer Churn Prediction

Year: 2023 | Conference Paper | Publisher: IEEE

10. A Customer Churn Prediction Model in Telecom Industry Using Boosting

Year: 2014 | Journal Article | Publisher: IEEE
 Reference
11. Machine Learning and Deep Learning Technique Used in Customer Churn Prediction: A Review
Year: 2023 | Conference Paper | Publisher: IEEE

12. A Comparison of Machine Learning Algorithms for Customer Churn Prediction

Year: 2023 | Conference Paper | Publisher: IEEE

13. Customer Churn Prediction in Telecommunication Industry Using Machine Learning and Deep
Learning Approach
Year: 2023 | Conference Paper | Publisher: IEEE

14. Profit Maximizing Logistic Regression Modeling for Customer Churn Prediction
Year: 2015 | Conference Paper | Publisher: IEEE

15. Churn Prediction With Explainability for the Customers of Telecom Industry
Year: 2023 | Conference Paper | Publisher: IEEE