Findings on Research Papers:

(Application of A.I)
PRESENTED BY: S ANJU KR.
SHRESTHA
01/06/2025
 Table of contents:
1. Nbias: A Natural Language Processing Framework for Bias Identification in Text

2. A decision support system for osteoporosis risk

3. Data-driven emergence of convolutional structure in neural networks

4. Navigating the data frontier in science assessment: Advancing data

augmentation strategies for machine learning applications with generative
artificial intelligence
5. Development of GUI for Text-to-Speech Recognition using Natural Language
Processing
Nbias: A Natural Language Processing Framework for Bias
Identification in Text
S.N. References Problem Statement Methodology Results
[1]. B. Hutchinson, V. • Data-related biases in • They Developed the Nbias • As compared to the
Prabhakaran, E. NLP repercussions framework with four layers: baseline model,
Denton, K. Webster, Y. unfair such mutations • Data gathering and corpus accuracy is
Zhong, S. Denuyl, or natural disapproval. construction. increased by up to
Social biases in NLP Either these models • Annotating bias-indicative terms 8%. Developed a
1 models as barriers for might have existed only using a semi-autonomous approach. robust domain-
persons with to cope with such an independent bias
disabilities issue on some fronts • Model development using BERT- detection
such as healthcare, based token classification for the framework for NLP
[2]. T. Bolukbasi, K.-W. social media, or entity "BIAS." applications.
Chang, J. Y. Zou, V. employment, but not • Evaluation combining qualitative and
Saligrama, A. T. Kalai, sufficiently able to give quantitative measures (F1 score, • Achieved an
Man is to computer generalized answers for robustness tests). accuracy
programmer as quite a long time. improvement of up
woman is to • They Used diverse datasets, to 8% against the
homemaker? including clinical, occupational, and baseline models.
debiasing word social media text. Proposed a
embeddings, framework for
Advances in neural robust, domain-
information agnostic bias
processing systems 29 detection in NLP
(2016). applications.

[3]. L. Dixon, J. Li, J.

 A decision support system for
S.N
osteoporosis
References
risk
Problem Statement Methodology Results
.
[1].https:// • Osteoporosis often • They Utilized an open-source dataset of 1. Achieved 89%
www.osteoporosis.f goes undiagnosed due 1493 patient records, employing 13 accuracy using a
oundation/facts- to its asymptomatic feature selection methods (e.g., Forward multi-level
statistics/ nature. This study Feature Selection, Grey Wolf Optimizer). ensemble learning
epidemiology-of- addresses the need for pipeline.
osteoporosis-and- a reliable system that • They Evaluated multiple machine
fragility-fractures, predicts osteoporosis learning models, including Random 2. Identified key
2 2023. (Accessed 15 risk and supports early Forest and Gradient Boosting, with predictive features
October 2023). intervention to ensemble stacking to identify optimal for osteoporosis,
improve outcomes. classifiers. such as bone
[2]. A.M. Parfitt, density and calcium
Bone remodeling • They Applied explainable AI techniques levels.
and bone loss: (e.g., SHAP, LIME) for interpretable
understanding the predictions. 3. Developed an
pathophysiology of explainable
osteoporosis, Clin. decision-support
Obstet. Gynecol. 30 system for clinical
(4) (Dec. 1987) 789– application.
811.

[3]. Steven,
“Osteoporosis Knee
X-Ray Dataset, 2022.
 Data-driven emergence of convolutional structure in
neural networks
S.N References Problem Statement Methodology Result
.
[1]. DiCarlo, J. J., • Previous work has • Developed a supervised 1. That non-Gaussian data
Zoccolan, D. & Rust, N. usually found that classification task using synthetic would induce local
C. How Does the Brain FCNNs do not self- datasets with translation invariance receptive fields in FCNNs.
Solve Visual Object organize into and non-Gaussian features.
Recognition? convolutional-like 2. Linked higher-order input
Neuron 73, 415–434 structures without • Conducted experiments with two- correlations to the
3 (2012). explicit layer neural networks to observe emergence of
regularization. pattern emergence. convolutional filters.
[2]. Yamins, D. L. K. et
al. Performance- • The goal of this • Analyzed network performance via 3. Provided insight into
optimized hierarchical contribution is to numerical simulations and tensor effective feature learning
models predict neural investigate decomposition techniques. both in artificial and
responses in conditions in biological systems.
higher visual cortex. which FCNNs
Proceedings of the develop localized,
National Academy of space-tiling
Sciences 111, 8619– receptive fields
8624. eprint: self-organizingly
https:// from data.
www.pnas.org/
content/
111/23/8619.full.pdf
(2014).
Navigating the data frontier in science
assessment
S.N. Reference Problem Statement Methodology Result
[1]. Alasadi, E. A., • Representative and • Comparing three different methods of 1. Indeed, combining
& Baiz, C. R. diversified training in data augmentation. the genuine
(2023). education confines the responses with the
Generative AI in performance of machine • Paraphrasing, by using AI chatbots' paraphrased
education and learning models to the tools for pre-existing responses. chatbot responses
research: formidable task of 2. worked best.
4 Opportunities, making accurate • Synthetic response generation by
concerns, and assessments of the subtle chatbots. Demonstrated
solutions. Journal reasoning carried out by enhanced
of Chemical students, especially when • Validation of improvements in ML scalability and
Education, facing higher-order performance based on a 24-category generalisability
100(8), categorization tasks such rubric of student reasoning. during formative
2965–2971. as the evaluation of 3. assessments.
https://doi.org/1 reaction mechanisms in • Assessment of ML-human score
0.1021/acs.jchem organic chemistry. agreement. Highlighted how AI
ed.3c00323
could decrease
[2].Anderson, P., demands on
Anson, C. M., resources when
Gonyea, R. M., & collecting data or
Paine, C. (2015). applying a rubric.
The contributions
of
writing to
 Development of GUI for Text-to-Speech Recognition using Natural
Language Processing

S.N References Problem Statement Methodology Results

.
[1]. J. Kanisha, G. • User-friendly text-to- • They created a TTS system • Synthesized speech
Balakrishanan, “Speech speech systems should be TTS Gramaty, based on that is high in
Transaction for Blinds developed for a better Natural Language Processing quality, allowing the
Using Speech-Text-Speech conversion facility from and Digital Signal Processing. user to have control
Conversions”, text to speech, making it Facilities for the user over the audio
Communications in more accessible to users interface such as the input of parameters.
5 Computer and Information who cannot see, as well as text and the synthesis of
Science book series (CCIS), giving the content an speech into MP3 files were • Enabled the storage
vol 131, part I, pp. 43-48, audio format. integrated. Built TTS engines of outputs from
2011 based on .NET Framework text-to-speech
3.5 using C#. processes as MP3
[2]. C. S. T. Thu, T. Zin, files for portability.
“Implementation of Text to
Speech Conversion”, • Issuing
International Journal of amendments for
Engineering Research & multilingual, live
Technology, vol. 3(3), pp. and web synthesis
911-915, 2014 usage in the future.

[3]. P. S. Shetake, S. A.
Patil, P. M. Jadhav, “Review
of Text To Speech
Conversion Methods”, 7