National Centre of Artificial Intelligence, University of the Punjab

The Governor of Punjab, in his capacity as the Chancellor of the University of Kamalia, has assigned additional charge of the Founding Dean, Faculty of Computing and Emerging Sciences, to Prof. Dr. Syed Waqar-ul-Qounain Jaffry Director National Centre of Artificial Intelligence, University of the Punjab. This appointment marks a significant milestone as the university expands its academic footprint in computing, artificial intelligence, and emerging technologies.

Prof. Dr. Syed Waqar ul Qounain Jaffry, director National Centre of Artificial Intelligence, University of the Punjab addresses Punjab University at 3rd NSPP Public Policy Conference on Artificial Intelligence in Public Policy and Governance

Lahore, August 25–26, 2025

The National School of Public Policy (NSPP), Lahore successfully hosted the 3rd Public Policy Conference on Artificial Intelligence in Public Policy and Governance on 25–26 August 2025. The two-day event brought together policymakers, researchers, academics, and practitioners to deliberate on the role of Artificial Intelligence (AI) in shaping Pakistan’s governance landscape.

The conference revolved around six key themes: AI and Socio-Economic Challenges, AI and Environmental Sustainability, AI in Public Sector Transformation, AI and Human Rights, Establishment of AI Valley, and AI and National Security Challenges.
Professor Dr. Syed Waqar ul Qounain Jaffry, Director, National Centre of Artificial Intelligence (NCAI) and Chairman, Department of Information Technology, University of the Punjab, participated as an invited discussant in the panel on AI and National Security Challenges.

During the panel and in his media interactions, Prof. Jaffry emphasized the critical role of AI in Pakistan’s counter-terrorism strategy, particularly through the use of surveillance, profiling, and predictive analytics. He highlighted how AI-powered tools are being globally deployed to enhance situational awareness, detect threats in real time, and support rapid response capabilities.
He appreciated the papers presented under the theme, noting their timely relevance in the context of the post-2021 resurgence of terrorism in Pakistan, especially in Balochistan and Khyber Pakhtunkhwa. He pointed out that the research offers comparative international insights, draws on Pakistan-specific case studies such as Safe City projects, and provides policy-oriented recommendations while also addressing ethical, legal, and civil liberty concerns.

At the same time, Prof. Jaffry acknowledged the need for deeper empirical research on Pakistan’s context, inclusion of local data-driven studies, and development of a robust AI legal and ethical framework aligned with international best practices. He encouraged scholars and practitioners to work on ethical AI models, localized datasets, and culturally aware AI training systems to ensure accuracy, fairness, and accountability in AI-based counter-terrorism applications.

He concluded that AI offers Pakistan a transformative opportunity in national security and governance, but its success will depend on responsible deployment, legal safeguards, and public trust.

The conference reaffirmed NSPP’s commitment to creating platforms for policy dialogue on emerging technologies and their integration into governance for sustainable national development.

چین کی قومی AI تعلیم پالیسی، 1 ستمبر 2025 سے نافذ العمل

چین کی وزارتِ تعلیم نے مئی 2025 میں بنیادی اور ثانوی سطح پر مصنوعی ذہانت (AI) کی تعلیم کو فروغ دینے کے لیے جامع قومی پالیسی اور رہنما اصول جاری کیے ہیں۔ یہ پالیسی 1 ستمبر 2025 سے نافذ ہوگی اور اس کا مقصد ایک تدریجی، جامع اور محفوظ نظام کے تحت طلبہ میں AI خواندگی، مہارت، اور تخلیقی صلاحیت پیدا کرنا ہے۔ اس پالیسی کے تحت چین نے 2030 تک مکمل قومی سطح پر AI تعلیم کو نصاب کا لازمی حصہ بنانے کا ہدف مقرر کیا ہے۔

پالیسی کے اہم پہلو درج ذیل ہیں:

لازمی AI تعلیم: یکم ستمبر 2025 سے، تمام پرائمری اور سیکنڈری اسکولوں میں سالانہ کم از کم آٹھ گھنٹے کی AI تعلیم دینا لازمی قرار دیا گیا ہے۔

تدریجی نصاب (Tiered Curriculum):

پرائمری اسکول: طلبہ کو بنیادی AI آگاہی دی جائے گی، جیسے آواز کی پہچان اور تصویر کی درجہ بندی۔
جونیئر ہائی اسکول: طلبہ AI منطق، مشین لرننگ اور تنقیدی سوچ پر مہارت حاصل کریں گے تاکہ جنریٹیو AI کے نتائج کو پرکھ سکیں۔
سینئر ہائی اسکول: طلبہ عملی جدت پر توجہ دیں گے، AI الگورتھمز ڈیزائن کریں گے، اور بین المضامینی سوچ پیدا کریں گے۔

اساتذہ کی تربیت: پالیسی میں اساتذہ کے تربیتی پروگراموں میں AI صلاحیتوں کو شامل کرنے کی ہدایت دی گئی ہے تاکہ وہ نئے نصاب کو مؤثر انداز میں پڑھا سکیں۔

اخلاقیات پر توجہ: طلبہ کو AI کے اخلاقی پہلوؤں جیسے الگورتھمک تعصب، پرائیویسی، اور سماجی ذمہ داری کی تربیت بھی دی جائے گی۔

موجودہ مضامین میں انضمام: اسکولوں کو ہدایت دی گئی ہے کہ وہ AI کو سائنس، IT اور دیگر مضامین کے ساتھ ساتھ بعد از اسکول پروگراموں میں بھی شامل کریں۔

جنریٹیو AI کے ضوابط: پرائمری طلبہ کو آزادانہ طور پر مواد تخلیق کرنے والے اوپن AI ٹولز استعمال کرنے سے روکا گیا ہے، جبکہ اساتذہ کو ہدایت دی گئی ہے کہ وہ تدریس، امتحانی سوالات یا طلبہ کے نتائج کے تجزیے کے لیے AI پر انحصار نہ کریں۔

مقاصد:

مستقبل کے لیے تیار ورک فورس پیدا کرنا جو صرف صارف نہیں بلکہ AI دنیا کے خالق، ڈویلپر اور قائد بن سکیں۔

قومی سطح پر 2030 تک AI تعلیم کی مکمل کوریج حاصل کرنا۔

طلبہ کو عملی مہارتوں سے آراستہ کر کے نئی تکنیکی ایجادات کی راہ ہموار کرنا۔

یہ پالیسی نہ صرف نصاب میں بنیادی تبدیلیاں لے کر آتی ہے بلکہ ایک محفوظ، شفاف، اور جامع تعلیمی ماڈل پیش کرتی ہے، جسے حکام عالمی سطح پر "چینی ماڈل آف AI ایجوکیشن" کے طور پر اجاگر کر رہے ہیں۔

China Issues National Policy to Promote AI Education in Schools (Effective September 1, 2025)

China’s Ministry of Education has announced a comprehensive national policy to promote artificial intelligence (AI) education across all primary and secondary schools, effective from September 1, 2025. The policy aims to integrate AI literacy and skills into the curriculum, with a goal of achieving nationwide coverage by 2030.

The initiative establishes a tiered, progressive, and spiral AI education system, moving from basic awareness in early grades to applied innovation at senior levels, while also ensuring the safe and ethical use of generative AI in classrooms.

Key Aspects of the Policy:

Mandatory AI Instruction:
From September 2025, all primary and secondary schools must provide at least eight hours of AI instruction annually.

Tiered Curriculum:

Primary School: Students are introduced to basic AI concepts such as voice recognition and image classification.

Junior High: Students learn about AI logic, machine learning, and how to critically evaluate generative AI outputs.

Senior High: The focus is on applied innovation, where students design and refine AI algorithms and develop interdisciplinary systems thinking.

Teacher Training:
AI-enabled teaching competencies will be included in teacher training programs to support the new curricula.

Ethics and Responsibility:
Students will be taught about AI ethics, algorithmic bias, data privacy, and social responsibility, ensuring that technical learning is balanced with human-centered values.

Integration with Existing Subjects:
Schools are encouraged to embed AI topics into existing subjects such as IT, science, and after-school programs, promoting project-based and hands-on learning.

Goals of the Policy:

Future-Ready Workforce: To prepare students not only as consumers but as creators, developers, and leaders in an AI-driven world.

National AI Literacy: To ensure full coverage of AI education nationwide by 2030.

Technological Innovation: To foster creativity and innovation by equipping students with the practical skills to design and develop AI systems.

Through this framework, China aims to build a safe, efficient, and inclusive AI-powered education system, offering a unique Chinese model for AI education that could influence global practices.

Prof. Dr. Syed Waqar ul Qounain Jaffry, Director National Centre of Artificial Intelligence (NCAI), University of the Punjab, Lahore, Delivers Lecture on on “Artificial Intelligence: Conceptual Understanding, Potential, and Applications” at NIPA Lahore

Lahore, August 22, 2025 — The National Institute of Public Administration (NIPA), Lahore, hosted Prof. Dr. Syed Waqar ul Qounain Jaffry, Chairman, Department of Information Technology and Director, National Centre of Artificial Intelligence, University of the Punjab, as a guest speaker in the 44th Mid-Career Management Course (MCMC).

Prof. Jaffry delivered an insightful lecture on “Artificial Intelligence: Conceptual Understanding, Potential, and Applications”, highlighting how ICT, data-driven governance, and AI are shaping the future of public sector management. The session covered a wide range of themes including machine learning, generative AI, ethical challenges, legal frameworks, and Pakistan’s newly approved National AI Policy 2025.

Drawing on international best practices and local success stories such as NADRA’s biometric systems, Punjab Safe Cities Authority’s AI surveillance, and the Ehsaas and Sehat Sahulat programs, Prof. Jaffry emphasized the critical role of AI in enhancing transparency, efficiency, and citizen-centered governance.

He urged officers to embrace ICT and AI-driven approaches in public service delivery, while ensuring ethical oversight, accountability, and inclusivity in their implementation. The session concluded with an engaging Q&A, where participants explored practical pathways for applying AI to improve governance in Pakistan.

Prof. Dr. Syed Waqar ul Qounain Jaffry, Director National Centre of Artificial Intelligence (NCAI), University of the Punjab, Lahore, represented Pakistan at IEEE IES SYP Congress 2025 in Tunisia.

Tunis, August 17, 2025 — Prof. Dr. Syed Waqar ul Qounain Jaffry, Director National Centre of Artificial Intelligence (NCAI), University of the Punjab, Lahore, proudly represented Pakistan at the 2nd IEEE Industrial Electronics Society (IES) Students and Young Professionals (SYP) Congress, held from 15–17 August 2025 at the Regency Hotel, Tunis, Tunisia. The international congress brought together over 120 participants from across the globe to exchange knowledge, build networks, and explore future directions in industrial electronics and emerging technologies.

Prof. Jaffry contributed a roundtable discussion on "Human Skills in an AI World: Preparing Future Engineers for Hybrid Intelligence." During the session, he emphasized the urgent need to equip young engineers with adaptive skills, ethical awareness, and interdisciplinary knowledge to thrive in the age of artificial intelligence. He further highlighted the importance of balancing technical expertise with human-centric competencies such as creativity, problem-solving, collaboration, emotional and cultural intelligence, conflict resolution, and leadership.

In his presentation, Prof. Jaffry also shared the current status, achievements, and progress of the IEEE Lahore Section, showcasing Pakistan’s growing contributions within the global IEEE community. His participation not only reflected Pakistan’s presence on the international stage but also inspired students and young professionals to prepare for the evolving landscape of technology with a holistic mindset.

The IEEE IES SYP Congress continues to serve as a premier platform for nurturing talent, inspiring leadership, and promoting global collaboration in advancing technological innovation for the benefit of humanity.

MPhil AI Thesis Defense: TD-MP-AI-12
Name: Javeria Rasool
Roll No: MSAIF22M002
Title: Machine Learning-Based Sleep Stage Classification: Analyzing Transitions and Distinctive Features Across Sleep Cycles
Supervisor: Dr. Muhammad Adeel Nisar
External Examiner: Dr. Wasim Ahmad
Date and Time: Friday, August 08, 2025, at 03:30 PM
Venue: AL-Khwarizmi Lecture Theater, Faculty of Computing and Information Technology, University of the Punjab, Old Campus, Lahore.
Abstract:
Classifying sleep stages is essential for comprehending sleep disorders and general health, but conventional approaches frequently have accuracy and generalization issues. By examining transitions and identifying commonalities throughout the five main stages of sleep, this work suggests a machine learning-based method for classifying sleep stages. Electroencephalogram (EEG) and electrooculogram (EOG) data were processed using polysomnography (PSG) recordings from several people in order to identify pertinent features that captured temporal and spectral properties. The best classification accuracy of 84.09% was obtained via an ensemble-based approach after many machine learning models were trained and assessed. Insights into sleep dynamics and their effects on overall sleep quality are provided by the study, which highlights significant changes between sleep stages. The outcomes demonstrate how well computational methods work to automate sleep analysis, with potential uses in clinical diagnoses and customized sleep tracking. By increasing classification accuracy and deepening our understanding of transitional behaviors between sleep phases,
this work advances the field of sleep research.We use comparative binary-pair analysis to show sensor- and feature-type dominance and identify and validate the most important transitional features for multi-stage sleep classification through systematic feature ranking.
Keywords: Sleep stage classification, Machine learning, Sleep transitions, XGBoost, Hand-crafted features, ISRUC-Sleep dataset, EOG, EEG, TSFEL, Sleep pattern analysis

MPhil AI Thesis Defense: TD-MP-AI-11
Name: Javed Iqbal
Roll No: MSAIF22M024
Title: Advancing Non-Invasive Interstitial Glucose Prediction Through Engineered Digital Biomarkers and Machine Learning
Supervisor: Dr. Muhammad Adeel Nisar
External Examiner: Dr. Wasim Ahmad
Date and Time: Friday, August 08, 2025, at 2:30 PM
Venue: AL-Khwarizmi Lecture Theater, Faculty of Computing and Information Technology, University of the Punjab, Old Campus, Lahore.
Abstract:
Glucose measuring is an important aspect of keeping metabolism healthy, i.e. prediabetic issue, type 2 diabetes. The devices with wearable functions together with continuous glucose monitoring (CGM) of glycemic variability are becoming increasingly popular due to the continuous data they provide. The primary goal of our research is to determine if wearable sensor technology can be utilized to identify prediabetes physiological changes in which sensors will be placed on the patient’s body to monitor the valuable health status. The present study employs data obtained from a group of subjects who possess Dexcom G6 CGM systems and Empatica E4 wristbands (wearable smart devices) during extended monitoring periods of time. The interstitial glucose gets captured by the Dexcom G6 CGM system, while E4 wristband from Empatica records photo plethysmography, electrodermal activity, and tri - axial accelerometry signals. We have extracted a total of 25 features from the data for our models.
We have utilized GRU and Bi-GRU to make analysis of gluconeogenesis and forecasting of glycemic extremes with the data collected from wearable technology devices. Our proposed models took advantage of multimodal inputs, and further through, supported architecture improvements to achieve performance efficiencies like MSE, MAPE, and RMSE. Moreover, we have tested our models for their effectiveness on data with and without food log data we examined to determine the accuracy of the outcome. GRU and Bi-GRU have got an average error MAPE (7.85%, 7.10%) respectively and without food log (8.95%, 8.06%), respectively. We have also performed analysis using the Clarke Error Grid (CEG) to analyze how many test points are lying in which zone. Based on the discussions above, general recurrent units (GRUs) can be used in modeling glycemic variability, and in the near future, this enhanced understanding of individual metabolic disorders would lead to more personalized healthcare.
We propose using wearable sensor technology, continuous glucose monitoring, and sophisticated machine learning techniques to design our study becoming a key breakthrough in personalized medicine and further positively affect people’s lives who are at increased risk of metabolic disorders.
Keywords: Glucose Prediction, Bi-GRU, GRU, Deep learning, Wearable devices, Personalized Model.

MPhil AI Thesis Defense: TD-MP-AI-10
Name: Rabia Hussain
Roll No: MSAIF22M016
Title: Stress and Anxiety Detection with Multimodal Sensor Data using Machine Learning, Deep Learning and Data Augmentation
Supervisor: Dr. Muhammad Adeel Nisar
External Examiner: Prof. Dr. Xinyu Huang, CTO Expand AI Luebeck. Germany
Date and Time: Friday, August 08, 2025, at 10:30 AM
Venue: AL-Khwarizmi Lecture Theater, Faculty of Computing and Information Technology, University of the Punjab, Old Campus, Lahore.
Abstract:
Physiological signal-based stress detection has become an essential field in affective computing, with growing application to mental health tracking and real-time well-being estimation. Wearable devices like RespiBAN and Empatica E4 provide ongoing multimodal physiological recordings, with signals such as EDA, ACC, and temperature being promising markers of stress responses. Despite increased research attention, difficulties remain in accurately modeling intricate temporal behavior and optimizing classification performance, particularly under subject-independent conditions, where generalizability is typically constrained. We have used the Wearable Stress and Affect Detection (WESAD) dataset consisting of multimodal physiological data collected using RespiBAN professional in the area of chest, and Empatica E4 detection on the wrist, on 15 participants. Features were extracted from both wrist and chest-worn devices, normalized, and augmented using BorderlineSMOTE to handle class imbalance. We utilized deep models like CNN-LSTM, CNN-BiLSTM, CNNGRU, and the Temporal Kolmogorov-Arnold Networks (TKAN) to model spatiotemporal relationships in physiological signals. Feature selection was carried out through Random Forest-based ranking to determine the top 25 features, improving model efficiency and interpretability. Both subject-dependent and Leave-One-Subject-Out (LOSO) cross-validation were implemented to see the effectiveness of personalized and generealized stress recogniution performance through model training. Experimental findings illustrated that the CNN-LSTM model was the most accurate with the maximum classification accuracy of 97.838 for subject-independent situations. Applying feature selection significantly enhanced learning curves, avoided overfitting, and produced stable F1-scores and ROC-AUC levels. Confusion matrices and ROC curves established that the models can effectively distinguish stress and non-stress classes under inter-subject variability as well. These results highlight the promise of integrating physiological signal analysis with deep learning and feature selection to support robust, real-time stress detection in wearable systems. The framework provides a scalable solution for future intelligent mental health monitoring system development and adds to the general understanding of stress classification across diverse user contexts.
Keywords: Stress detection, Anxiety detection, WESAD dataset, Wearable sensors, Physiological signal classification, CNN-BiLSTM, TKAN, Feature selection, BSMOTE, Deep learning, Affective computing, Multimodal data fusion, Real-time monitoring, Biomedical signal processing, Lightweight models.

MPhil AI Thesis Defense: TD-MP-AI-09
Name: Syed Muhammad Hamza
Roll No: MSAIF22M010
Title: Optimizing Sleep Stage Classification with Deep Learning Using Sensor Data Augmentation Strategy
Supervisor: Dr. Muhammad Adeel Nisar
External Examiner: Dr. Ayesha Atta
Date and Time: Friday, August 08, 2025, at 09:30 AM
Venue: AL-Khwarizmi Lecture Theater, Faculty of Computing and Information Technology, University of the Punjab, Old Campus, Lahore.
Abstract:
Accurately identifying sleep stages is critical for improved health management and clinical diagnosis, especially when data-driven approaches become more prominent. Sensor technology advancements have enabled increased detail and intrusion-free analysis of sleep patterns, resulting in previously unattainable insights. Traditional methods for categorizing sleep stages have mostly relied on labor-intensive, human error-prone manual feature extraction and signal processing techniques, despite their effectiveness. However, the emergence of deep learning represents a paradigm shift in how we approach this challenge. Previous studies have employed various deep learning architectures such as CNNs and RNNs to classify sleep stages, with mixed results in terms of accuracy and generalization. Since deep learning models are prone to overfitting due to the training on smaller relative labeled dataset and as well due to the complexity of their structure, dropout and batch normalization are some of the methods designed to solve these problems. The overfitting is avoided due to dropout and batch normalization that accelerates the model convergence yielding
stronger and trusty classification outcomes. In our thesis, we are going to discuss how deep learning can enhance sleep stage classification through multimodal biosignals, namely EEG, EOG, EMG, of the ISRUC Sleep dataset (Group 1) containing a detailed record of 100 participants. The experimental findings show high performance in terms of accuracy, precision, recall, F1-score and Cohen Kappa coefficient, which are measured in various parts of the dataset. The fact that the results demonstrate that deep learning should be used in sleep studies more widely is remarkable because the suggested model demonstrates better classification and generalization. These outcomes suggest more advanced, automated, and scalable sleep staging practice, which has at least substantial value in terms of sleep disorders diagnosis and the creation of data-driven and efficient health tracking systems. This has substantial implications for future sleep research. It may improve approaches for
identifying sleep disorders and contribute to the development of data-driven, more efficient health management systems. The study reported here represents a significant development in the ongoing effort to apply cutting-edge machine learning approaches to improve human health.
Keywords: Deep Learning, Sensor Data Augmentation, Polysomnography (PSG), Electroencephalogram (EEG), Electrooculogram (EOG), Electromyogram (EMG)