Interview with Zahra Ghorrati: creating frameworks for human exercise recognition utilizing wearable sensors

On this interview sequence, we’re assembly a number of the AAAI/SIGAI Doctoral Consortium individuals to search out out extra about their analysis. Zahra Ghorrati is creating frameworks for human exercise recognition utilizing wearable sensors. We caught up with Zahra to search out out extra about this analysis, the points she has discovered most fascinating, and her recommendation for potential PhD college students.

Inform us a bit about your PhD – the place are you finding out, and what’s the matter of your analysis?

I’m pursuing my PhD at Purdue College, the place my dissertation focuses on creating scalable and adaptive deep studying frameworks for human exercise recognition (HAR) utilizing wearable sensors. I used to be drawn to this matter as a result of wearables have the potential to remodel fields like healthcare, aged care, and long-term exercise monitoring. Not like video-based recognition, which might elevate privateness considerations and requires fastened digital camera setups, wearables are transportable, non-intrusive, and able to steady monitoring, making them ideally suited for capturing exercise knowledge in pure, real-world settings.

The central problem my dissertation addresses is that wearable knowledge is commonly noisy, inconsistent, and unsure, relying on sensor placement, motion artifacts, and system limitations. My objective is to design deep studying fashions that aren’t solely computationally environment friendly and interpretable but in addition strong to the variability of real-world knowledge. In doing so, I intention to make sure that wearable HAR programs are each sensible and reliable for deployment exterior managed lab environments.

This analysis has been supported by the Polytechnic Summer time Analysis Grant at Purdue. Past my dissertation work, I contribute to the analysis group as a reviewer for conferences corresponding to CoDIT, CTDIAC, and IRC, and I’ve been invited to overview for AAAI 2026. I used to be additionally concerned in group constructing, serving as Native Organizer and Security Chair for the twenty fourth Worldwide Convention on Autonomous Brokers and Multiagent Techniques (AAMAS 2025), and persevering with as Security Chair for AAMAS 2026.

Might you give us an outline of the analysis you’ve carried out up to now throughout your PhD?

Thus far, my analysis has targeted on creating a hierarchical fuzzy deep neural community that may adapt to various human exercise recognition datasets. In my preliminary work, I explored a hierarchical recognition method, the place less complicated actions are detected at earlier ranges of the mannequin and extra advanced actions are acknowledged at greater ranges. To reinforce each robustness and interpretability, I built-in fuzzy logic rules into deep studying, permitting the mannequin to higher deal with uncertainty in real-world sensor knowledge.

A key power of this mannequin is its simplicity and low computational price, which makes it significantly nicely fitted to real-time exercise recognition on wearable units. I’ve rigorously evaluated the framework on a number of benchmark datasets of multivariate time sequence and systematically in contrast its efficiency in opposition to state-of-the-art strategies, the place it has demonstrated each aggressive accuracy and improved interpretability.

Is there a facet of your analysis that has been significantly fascinating?

Sure, what excites me most is discovering how totally different approaches could make human exercise recognition each smarter and extra sensible. As an example, integrating fuzzy logic has been fascinating, as a result of it permits the mannequin to seize the pure uncertainty and variability of human motion. As an alternative of forcing inflexible classifications, the system can motive by way of levels of confidence, making it extra interpretable and nearer to how people really assume.

I additionally discover the hierarchical design of my mannequin significantly fascinating. Recognizing easy actions first, after which constructing towards extra advanced behaviors, mirrors the best way people typically perceive actions in layers. This construction not solely makes the mannequin environment friendly but in addition supplies insights into how totally different actions relate to 1 one other.

Past methodology, what motivates me is the real-world potential. The truth that these fashions can run effectively on wearables means they might ultimately help personalised healthcare, aged care, and long run exercise monitoring in folks’s on a regular basis lives. And because the methods I’m creating apply broadly to time sequence knowledge, their affect might prolong nicely past HAR, into areas like medical diagnostics, IoT monitoring, and even audio recognition. That sense of each depth and flexibility is what makes the analysis particularly rewarding for me.

What are your plans for constructing in your analysis up to now throughout the PhD – what points will you be investigating subsequent?

Shifting ahead, I plan to additional improve the scalability and adaptableness of my framework in order that it might probably successfully deal with giant scale datasets and help real-time functions. A significant focus might be on enhancing each the computational effectivity and interpretability of the mannequin, making certain it isn’t solely highly effective but in addition sensible for deployment in real-world situations.

Whereas my present analysis has targeted on human exercise recognition, I’m excited to broaden the scope to the broader area of time sequence classification. I see nice potential in making use of my framework to areas corresponding to sound classification, physiological sign evaluation, and different time-dependent domains. This may enable me to reveal the generalizability and robustness of my method throughout various functions the place time-based knowledge is vital.

In the long run, my objective is to develop a unified, scalable mannequin for time sequence evaluation that balances adaptability, interpretability, and effectivity. I hope such a framework can function a basis for advancing not solely HAR but in addition a broad vary of healthcare, environmental, and AI-driven functions that require real-time, data-driven decision-making.

What made you need to research AI, and specifically the realm of wearables?

My curiosity in wearables started throughout my time in Paris, the place I used to be first launched to the potential of sensor-based monitoring in healthcare. I used to be instantly drawn to how discreet and non-invasive wearables are in comparison with video-based strategies, particularly for functions like aged care and affected person monitoring.

Extra broadly, I’ve at all times been fascinated by AI’s potential to interpret advanced knowledge and uncover significant patterns that may improve human well-being. Wearables supplied the right intersection of my pursuits, combining cutting-edge AI methods with sensible, real-world affect, which naturally led me to focus my analysis on this space.

What recommendation would you give to somebody considering of doing a PhD within the subject?

A PhD in AI calls for each technical experience and resilience. My recommendation could be:

• Keep curious and adaptable, as a result of analysis instructions evolve shortly, and the power to pivot or discover new concepts is invaluable.
• Examine combining disciplines. AI advantages vastly from insights in fields like psychology, healthcare, and human-computer interplay.
• Most significantly, select an issue you might be really keen about. That zeal will maintain you thru the inevitable challenges and setbacks of the PhD journey.

Approaching your analysis with curiosity, openness, and real curiosity could make the PhD not only a problem, however a deeply rewarding expertise.

Might you inform us an fascinating (non-AI associated) reality about you?

Exterior of analysis, I’m keen about management and group constructing. As president of the Purdue Tango Membership, I grew the group from simply 2 college students to over 40 lively members, organized weekly courses, and hosted giant occasions with internationally acknowledged instructors. Extra importantly, I targeted on making a welcoming group the place college students really feel related and supported. For me, tango is greater than dance, it’s a option to convey folks collectively, bridge cultures, and steadiness the depth of analysis with creativity and pleasure.

I additionally apply these expertise in tutorial management. For instance, I function Native Organizer and Security Chair for the AAMAS 2025 and 2026 conferences, which has given me hands-on expertise managing occasions, coordinating groups, and creating inclusive areas for researchers worldwide.

About Zahra

Zahra Ghorrati is a PhD candidate and instructing assistant at Purdue College, specializing in synthetic intelligence and time sequence classification with functions in human exercise recognition. She earned her undergraduate diploma in Pc Software program Engineering and her grasp’s diploma in Synthetic Intelligence. Her analysis focuses on creating scalable and interpretable fuzzy deep studying fashions for wearable sensor knowledge. She has offered her work at main worldwide conferences and journals, together with AAMAS, PAAMS, FUZZ-IEEE, IEEE Entry, System and Utilized Gentle Computing. She has served as a reviewer for CoDIT, CTDIAC, and IRC, and has been invited to overview for AAAI 2026. Zahra additionally contributes to group constructing as Native Organizer and Security Chair for AAMAS 2025 and 2026.