Page 152 - Kaleidoscope Academic Conference Proceedings 2024

2024 ITU Kaleidoscope Academic Conference
smartphone utilization on learning outcomes. However, a research neural network research, influencing the trajectory of artificial
gap emerges regarding dedicated educational applications lever- intelligence research and application development.
aging advanced technologies, such as OCR and deep learning.
E. A Novel Connectionist System for Unconstrained Handwriting
While existing literature focuses broadly on smartphone use, there
Recognition
is limited exploration into tailored applications for foundational
education. The Alpha-Bit initiative aims to address this gap The research paper by A. Graves et al., titled ”A novel
by introducing an Android application that utilizes OCR and connectionist system for unconstrained handwriting recognition,”
deep learning for targeted and accessible early childhood educa- published in the IEEE Transactions on Pattern Analysis and
tion, contributing a unique perspective to the ongoing discourse Machine Intelligence in 2009, represents a significant milestone
on smartphone use and academic effectiveness among primary in the realm of handwriting recognition. [2] Against the backdrop
school students. of an increasing need for systems capable of deciphering uncon-
strained and diverse handwritten input, Graves and his co-authors
C. Teaching Mathematics with Mobile Devices and the Realistic introduce a novel connectionist approach. Their work addresses
Mathematical Education (RME) Approach in Kindergarten the inherent challenges associated with variability in writing
The research paper titled ”Teaching Mathematics with Mobile styles, character shapes, and contextual dependencies present in
Devices and the Realistic Mathematical Education (RME) Ap- unconstrained handwriting. By proposing a connectionist system,
proach in Kindergarten” provides a comprehensive exploration the authors leverage the power of recurrent neural networks
of integrating mobile devices into the teaching of mathematics (RNNs), enabling the model to capture temporal dependencies in
for kindergarten students, utilizing the Realistic Mathematical the handwriting sequence. The paper provides a detailed exposi-
Education (RME) approach. [8] Executed at the Department of tion of the architecture, training methodology, and performance
Preschool Education, University of Crete, the study strategically evaluation of the proposed system. This contribution is pivotal in
incorporates tablet-type devices to implement RME across three advancing the field, as it not only introduces a new model but
levels, focusing on foundational mathematical concepts tailored also offers insights into the potential of recurrent neural networks
for the kindergarten level. The findings underscore the effective- for handwriting recognition tasks. The study’s impact extends
ness of well-designed digital educational activities in enhancing beyond its immediate context, influencing subsequent research in
learning experiences, enabling children to acquire knowledge the design of neural network architectures for pattern recognition
through activities aligned with their immediate interests and real- in unconstrained handwritten text, thereby contributing to the
life scenarios, particularly in the realm of mathematics. Linking ongoing evolution of machine learning techniques for real-world
this research to the Alpha-Bit initiative, the study emphasizes applications.
the transformative potential of mobile devices in early childhood
F. SEED: Semantics Enhanced Encoder-Decoder Framework for
education, particularly for foundational subjects like mathematics.
Scene Text Recognition
However, a research gap becomes evident concerning Alpha-
Bit’s specific focus on leveraging Optical Character Recognition The research paper authored by T. Qiao et al., titled ”Seed:
(OCR) and deep learning for alphabets and numbers. While Semantics enhanced encoder-decoder framework for scene text
existing literature explores the broader benefits of mobile devices recognition,” and presented in the Proceedings of the IEEE/CVF
in education, there is a distinct lack of dedicated exploration Conference on Computer Vision and Pattern Recognition in
into mobile applications employing advanced OCR technolo- 2020, makes a significant contribution to the field of scene
gies for foundational learning. Alpha-Bit aims to fill this gap text recognition. [3] In the context of computer vision and
by introducing an innovative Android application that utilizes pattern recognition, scene text recognition poses a unique set
OCR and deep learning, providing targeted and accessible early of challenges due to variations in fonts, sizes, and orientations
childhood education, thereby contributing a unique perspective of text in natural images. Qiao and colleagues address these
to the landscape of mobile-assisted learning for foundational challenges by proposing a novel framework, ”Seed,” that inte-
mathematics and literacy. grates semantics to enhance the capabilities of an encoder-decoder
architecture. The paper meticulously details the design of Seed,
D. Backpropagation Applied to Handwritten Zip Code Recogni-
emphasizing the semantic information incorporation process and
tion
the resulting improvements in scene text recognition performance.
Yann LeCun et al.’s seminal work, ”Backpropagation Applied By leveraging the advancements in encoder-decoder architectures
to Handwritten Zip Code Recognition,” published in Neural and introducing semantic enrichment, the authors showcase how
Computation in 1989, marks a pivotal contribution to the field of Seed outperforms existing methods in accurately recognizing text
neural networks and pattern recognition. [1] This paper lays the within complex visual scenes. This work not only contributes
foundation for the application of backpropagation, a fundamental a specific solution to the challenges of scene text recognition
neural network training algorithm, to the domain of handwritten but also reflects a broader trend in the field, highlighting the
zip code recognition. The authors delve into the complexities of increasing importance of semantic understanding in improving the
training neural networks to recognize diverse handwriting styles, accuracy and robustness of computer vision systems. The findings
demonstrating the efficacy of backpropagation in optimizing the presented by Qiao et al. contribute to the evolving landscape of
network’s parameters. The research is situated within the broader scene text recognition, with implications for applications in image
context of the evolving field of artificial intelligence, specifically understanding, document analysis, and other domains reliant on
addressing the challenges associated with optical character recog- effective text extraction from complex visual data.
nition. By focusing on handwritten zip code recognition, LeCun et
al. not only contribute a practical application but also advance the G. Tencent ML-Images: A Large-Scale Multi-Label Image
theoretical understanding of how neural networks can be trained Database for Visual Representation Learning
effectively for pattern recognition tasks. The paper’s enduring The research paper authored by Z. Wan et al., titled ”Tencent
significance lies in its pioneering role in the development of ML-Images: A Large-Scale Multi-Label Image Database for
deep learning methodologies, setting the stage for subsequent Visual Representation Learning,” and published in IEEE Access
breakthroughs in image recognition and machine learning. LeCun in 2019, contributes significantly to the field of computer vision
et al.’s work serves as a cornerstone in the rich history of and visual representation learning. [4] In response to the growing
– 108 –