codesanitize

In a collection of interviews, we’re assembly a few of the AAAI/SIGAI Doctoral Consortium contributors to seek out out extra about their analysis. On this newest interview, we hear from Amina Mević who’s making use of machine studying to semiconductor manufacturing. Discover out extra about her PhD analysis to date, what makes this subject so attention-grabbing, and the way she discovered the AAAI Doctoral Consortium expertise.

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

I’m at the moment pursuing my PhD on the College of Sarajevo, School of Electrical Engineering, Division of Laptop Science and Informatics. My analysis is being carried out in collaboration with Infineon Applied sciences Austria as a part of the Essential Undertaking of Frequent European Curiosity (IPCEI) in Microelectronics. The subject of my analysis focuses on creating an explainable multi-output digital metrology system primarily based on machine studying to foretell the bodily properties of metallic layers in semiconductor manufacturing.

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

Within the first 12 months of my PhD, I labored on preprocessing complicated manufacturing information and making ready a strong multi-output prediction setup for digital metrology. I collaborated with trade consultants to grasp the method intricacies and validate the prediction fashions. I utilized a projection-based choice algorithm (ProjSe), which aligned effectively with each area information and course of physics.

Within the second 12 months, I developed an explanatory technique, designed to determine probably the most related enter options for multi-output predictions.

Is there a side of your analysis that has been notably attention-grabbing?

For me, probably the most attention-grabbing side is the synergy between physics, arithmetic, cutting-edge expertise, psychology, and ethics. I’m working with information collected throughout a bodily course of—bodily vapor deposition—utilizing ideas from geometry and algebra, notably projection operators and their algebra, which have roots in quantum mechanics, to boost each the efficiency and interpretability of machine studying fashions. Collaborating intently with engineers within the semiconductor trade has additionally been eye-opening, particularly seeing how explanations can instantly assist human decision-making in high-stakes environments. I really feel really honored to deepen my information throughout these fields and to conduct this multidisciplinary analysis.

What are your plans for constructing in your analysis to date through the PhD – what elements will you be investigating subsequent?

I plan to focus extra on time collection information and develop explanatory strategies for multivariate time collection fashions. Moreover, I intend to analyze elements of accountable AI throughout the semiconductor trade and be sure that the options proposed throughout my PhD align with the rules outlined within the EU AI Act.

How was the AAAI Doctoral Consortium, and the AAAI convention expertise generally?

Attending the AAAI Doctoral Consortium was a tremendous expertise! It gave me the chance to current my analysis and obtain beneficial suggestions from main AI researchers. The networking side was equally rewarding—I had inspiring conversations with fellow PhD college students and mentors from world wide. The principle convention itself was energizing and numerous, with cutting-edge analysis offered throughout so many AI subfields. It undoubtedly strengthened my motivation and gave me new concepts for the ultimate part of my PhD.

Amina presenting two posters at AAAI 2025.

What made you wish to examine AI?

After graduating in theoretical physics, I discovered that job alternatives—particularly in physics analysis—had been fairly restricted in my nation. I started searching for roles the place I may apply the mathematical information and problem-solving abilities I had developed throughout my research. On the time, information science seemed to be a really perfect and promising subject. Nevertheless, I quickly realized that I missed the depth and objective of basic analysis, which was typically missing in trade roles. That motivated me to pursue a PhD in AI, aiming to realize a deep, foundational understanding of the expertise—one that may be utilized meaningfully and utilized in service of humanity.

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

Keep curious and open to studying from completely different disciplines—particularly arithmetic, statistics, and area information. Be sure that your analysis has a objective that resonates with you personally, as that zeal will assist carry you thru challenges. There might be moments while you’ll really feel like giving up, however earlier than making any choice, ask your self: am I simply drained? Typically, relaxation is the answer to a lot of our issues. Lastly, discover mentors and communities to share concepts with and keep impressed.

Might you inform us an attention-grabbing (non-AI associated) truth about you?

I’m an enormous science outreach fanatic! I often volunteer with the Affiliation for the Development of Science and Expertise in Bosnia, the place we run workshops and occasions to encourage children and highschool college students to discover STEM—particularly in underserved communities.

About Amina

Amina Mević is a PhD candidate and educating assistant on the College of Sarajevo, School of Electrical Engineering, Bosnia and Herzegovina. Her analysis is carried out in collaboration with Infineon Applied sciences Austria as a part of the IPCEI in Microelectronics. She earned a grasp’s diploma in theoretical physics and was awarded two Golden Badges of the College of Sarajevo for attaining a GPA increased than 9.5/10 throughout each her bachelor’s and grasp’s research. Amina actively volunteers to advertise STEM schooling amongst youth in Bosnia and Herzegovina and is devoted to enhancing the analysis surroundings in her nation.

The AI Gold Rush – From Pilots and Experiments to Enterprise Scale and Technique

Moore’s Legislation is nicely and really in play on the subject of AI. AI is closely in demand, and each enterprise is adopting AI. Innovation can be serving to gasoline this demand with new AI fashions, AI Brokers, and new applied sciences coming into this place. That is making a elementary shift for enterprises – the stage for pilots and funky experiments and showcases for AI, specifically, Generative AI is essentially fading. Enterprises are realizing that AI must be embedded as a part of the Enterprise technique for scaling and creating true enterprise differentiation. AI is a subject in most boardrooms, leading to strategic innovation and budgets.

Knowledge: The First Domino in AI Technique

A key consideration in any AI technique must be Knowledge. Knowledge is crucial for AI fashions to be contextual, clever, and area and enterprise-specific. AI fashions predict outcomes based mostly on each the way in which the mannequin is tuned and the inputs introduced to it. Each of those depend upon the standard, selection, recency, and construction of the information.

Based on a current IDC forecast, AI is anticipated to spice up the worldwide financial system by almost $20 trillion by 2030, pushed not solely by fashions but in addition by huge investments within the underlying information and infrastructure that gasoline them.

Coaching information with slender subsets results in biased fashions, outdated information results in irrelevant outcomes, and poor information simply results in poor AI outcomes. Subsequently, Knowledge is the primary domino in an enterprise’s information technique. Even with one of the best folks and cutting-edge applied sciences, if the information domino falls, your entire AI technique tumbles down rapidly.

As Gartner’s 2024 report on high information and analytics developments notes, organizations as they scale with AI depend upon information, and the leaders who succeed shall be those that set up belief of their information and lead with it strategically.

Key Strategic Knowledge Choices on your AI Technique

Listed here are 5 key issues you and your enterprise must make for on making ready your Knowledge on your AI technique:

1.  Reuse your Knowledge panorama – A number of enterprises don’t reuse the information administration, information governance, and information storage and analytics panorama for AI. Loads of information serving crucial reporting and analytics will also be crucial for AI. It’s due to this fact essential to start out with the information belongings already current within the enterprise. After all, this must be augmented with the precise information high quality measures.

Key Query to Ask – What information do we have now in our enterprise, and what situation is it in?

2.  Metadata and Knowledge Lineage – For the information in place, metadata, i.e., information concerning the information, is perhaps simply as crucial, if no more, for AI. As an illustration, the enterprise phrases tagged to the information may also help determine the related context for a RAG mannequin, as an example. When a person asks for the standing of a declare in an Insurance coverage enterprise, all the information attributes tagged with Declare standing can be utilized as context for the AI mannequin to reply. Knowledge Lineage additionally helps perceive the move of the information, serving to the AI fashions to determine trusted information sources.

Based mostly on a current ISASA weblog, AI Governance is crucial and requires the precise metadata and information lineage to scale.

Key Query to Ask – Is our information tagged correctly with enterprise and technical metadata? Can we acquire information lineage to grasp how the information flows finish to finish?

3.  Knowledge Governance and Compliance – Be certain that your information is nicely ruled and managed, and that any compliance and privateness rules are utilized to the information. The AI Technique ought to then inherit and prolong these governance and rules than ranging from scratch. As an illustration, if a buyer desires their information to be anonymized as per GDPR rules, an AI mannequin must be each skilled and operational on the anonymized dataset.

Key Query to Ask – Do we have now a Knowledge Governance and Compliance program in place? If not, what are the important thing facets that I must have in place for my AI technique?

4.  Deal with Grasp Knowledge as your AI Quarterback – Essential Grasp Knowledge, which comprises information about the important thing entities in your enterprise, must be used as the bottom on your AI technique. As an illustration, if the 360 diploma view of a buyer exists, an AI technique on any buyer area, equivalent to a buyer churn prediction, ought to leverage this grasp information to keep away from any information missed or incomplete. After all, this may be mixed with extra data from particular information sources.

Key Query to Ask – Do I’ve my crucial grasp information domains accessible in a whole and related to the remainder of my information panorama?

5.  Knowledge and its worth – Knowledge shouldn’t be handled as a value heart however measured by way of its worth, each in direction of AI and the enterprise. This requires information to be on Board and CXO matters along with AI.

Key Query to Ask – Does my Board and CXOs perceive the worth of Knowledge to the group? If not, how can we be sure that that is understood, particularly within the context of the AI technique within the enterprise?

Fashions Come and Go, However Knowledge Endures.

As your AI technique evolves, new fashions and AI improvements will emerge. The pace of innovation on this house is mind-boggling. However over time, AI fashions will commoditize; the true differentiator in your enterprise just isn’t which mannequin you utilize however the way it will get contextualized with what information is coaching, fine-tuning, and dealing on it.

In case you’re crafting an AI technique, don’t begin with the mannequin. Begin with the query: Do we have now the information to assist it?