Advancing Research Frontiers
Advancing Research Frontiers
This page presents my peer-reviewed publications and ongoing research focused on machine learning, signal processing, and intelligent human-centered systems. My work emphasizes the development of robust, generalizable models for real-world data, particularly in domains involving speech, physiological signals, and complex time-series systems.
My research approach integrates deep learning, statistical modeling, and feature engineering, with a strong focus on:
cross-dataset generalization
multimodal learning
reproducibility and evaluation rigor
These contributions aim to bridge the gap between theoretical AI models and deployable systems, particularly in affective computing, neurotechnology, and intelligent analytics.
Conference Publication:
Ahmed, W., Riaz, S., Iftikhar, K., Konur, S. (2023). Speech Emotion Recognition Using Deep Learning.
In: Bramer, M., Stahl, F. (eds) Artificial Intelligence XL. SGAI 2023.
Lecture Notes in Computer Science (), vol 14381. Springer, Cham.
https://doi.org/10.1007/978-3-031-47994-6_14
[ Summary ]
Speech Emotion Recognition (SER) remains a challenging problem due to cross-speaker variability, dataset bias, and limited generalization across domains.
In this work, we propose a deep learning-based SER framework designed to improve robustness through multi-dataset integration. Four benchmark datasets (RAVDESS, TESS, CREMA-D, SAVEE) were unified to construct a diverse and heterogeneous training corpus, addressing limitations of single-dataset learning.
A comprehensive feature extraction pipeline was implemented, incorporating:
MFCCs
Mel Spectrograms
Chroma Features
Zero Crossing Rate (ZCR)
Root Mean Square Energy (RMS)
These features were used to train a multi-layer Convolutional Neural Network (CNN) for emotion classification.
The model achieved 76% accuracy, demonstrating improved generalization compared to baseline single-dataset approaches.
This work highlights the importance of data diversity and feature representation in SER systems and provides a foundation for future research in:
domain adaptation
multimodal emotion recognition
real-world deployment of affective AI systems
–
Journal Publication:
Gondal, Taimoor & Iftikhar, Khunsa & Hayyat, Umar. (2017).
CODEN(USA): A novel inverter design with optimized multilevel programmed drive sequences.
The Journal of Scientific and Engineering Research. 4. 35-42.
https://jsaer.com/archive/volume-4-issue-6-2017/
[ Summary ]
This paper presents a novel multilevel inverter topology designed to improve efficiency and reduce harmonic distortion while minimizing system complexity. A four-level inverter architecture is introduced, utilizing a circuit topology that requires only a single isolation supply for driving all MOSFET gate circuits.
The proposed multilevel programmed drive sequences are an extension of binary programmed pulse-width modulation, enabling a broader set of voltage input levels. These sequences are optimized using Mixed Integer Linear Programming (MILP) implemented in MATLAB. The optimized sequences are validated through circuit simulations conducted in PSPICE using the SPLS interface.
Comparative harmonic analysis against conventional binary inverters demonstrates the superior performance of the proposed approach in terms of harmonic reduction and signal quality, validating its effectiveness for power electronics and energy systems applications.
Keywords: Multilevel Inverter, PWM Optimization, Mixed Integer Linear Programming, Power Electronics, Harmonic Analysis
Manuscript in Preparation
EEG-Based Cognitive State Modelling and Attention Training using Brain-Computer Interfaces
Status: In preparation
Overview:
This research focuses on developing a BCI-based framework for modeling attention and cognitive states using EEG signals, with applications in neurodevelopmental conditions such as ADHD.
The work extends prior implementation of a BCI-based attention training system and advances it toward a research-grade experimental framework.
Key Contributions:
EEG signal acquisition and preprocessing using Emotiv platforms
Feature extraction and analysis of cognitive state indicators
Development of interactive neurofeedback-based training environments
Session-based evaluation across users to analyze attention modulation
Research Direction :
This work aims to contribute toward:
data-driven cognitive state modelling
adaptive neurofeedback systems
integration of machine learning with EEG-based human–computer interaction
The long-term goal is to develop scalable and clinically relevant BCI systems, suitable for publication in venues related to:
computational neuroscience
affective computing
human-centered AI
Note: Further publications will be added as ongoing research progresses and manuscripts move through the peer-review process.
Disclaimer: The publications on this website belong to Springer, Semantic Scholar, ResearchGate, JSAER, Elsevier, IEEE, etc. Copyright is retained by the authors or other holders. Personal use is permitted, but for other uses such as reprinting or republishing for promotional purposes, permission must be obtained from the respective publishers. Unauthorized reproduction, distribution, or modification is prohibited.