Notre Dame Data Science’s Post

As the same NYT reported in June this year 'Everyone Is Using A.I. for Everything. Is That Bad?'. Yes. Yes it is. And as the AI bubble grows and people feel more and more like they have to be part of it - the voices saying that the ways in which AI adoption is leeching into every corner of people's lives is deleterious seem fewer and fewer. But when we allow AI to do everything for us - it is not hard to realise that we start to be able to be do it for ourselves less. Of course there can be benign and transformative uses of AI - *but* if people rely on it for all tasks that would otherwise require thought, comprehension, and expression - we will see a dangerous diminishment of the capability of people to accomplish those tasks. https://lnkd.in/eqfU-6DY

Behind every great AI system is solid math – and understanding it unlocks everything else. At ODSC AI West, Dr. David Hoyle, Research Data Science Specialist at dunnhumby and author of 15 Math Concepts Every Data Scientist Should Know, will lead a foundational bootcamp: Introduction to Math for Data Science. With a background spanning academia and industry – from Associate Professor of Machine Learning to building global demand forecasting models – David makes complex math concepts accessible and applicable. In this session, you’ll: – Strengthen your understanding of core mathematical principles in data science – Connect theory to real-world machine learning and AI applications – Learn how math drives better modeling, inference, and interpretation – Build confidence applying these concepts across your own projects 🔗 Register now → https://hubs.li/Q03Gqg1J0 #ODSCAI #DataScience #MachineLearning #AI #MathForDataScience #ODSC

AI and Academic Integrity has been at the forefront of educators minds this year. At the University of Illinois Urbana-Champaign, two professors discovered that a significant number of students used AI to write their apology email when the emails all started with the same phrase. While an apology email isn't a high stakes exam or something students were explicitly told not to use AI on, it does prove that AI is becoming ingrained in students experience. Institutions and educators are struggling to keep up with changing technology and are often at a loss on how to handle AI dishonesty. If you or your institution doesn't have a policy that address AI, the time to create one is now. Read the full The New York Times article here: https://lnkd.in/eiaMxC-E

📊 Corelia™, Part VIII: Robust Statistics and Soft Computing - Strength in Uncertainty In complex, real-world environments, data is noisy, incomplete, and often misleading. Corelia embraces uncertainty - not as a weakness - but as a fundamental reality. To navigate this, she relies on robust statistics and soft computing methods that emphasize resilience, adaptability, and grace under pressure. 📈 Why Robust Statistics? Traditional methods like arithmetic means can be fragile in the face of outliers or skewed data. Corelia prefers robust statistics - such as medians, trimmed means, and quantile-based measures - which better capture central tendencies without being thrown off by anomalies. This makes her: - More resistant to noise and deceptive inputs - Better at identifying meaningful patterns - Less prone to catastrophic failures triggered by rare events 🌱 The Role of Soft Computing Soft computing refers to techniques inspired by human reasoning and natural processes - including: - Fuzzy logic, allowing Corelia to handle ambiguous or partial truths - Rough logic, allowing Corelia to handles uncertainty and incomplete data - Neural-inspired networks, providing pattern recognition and generalization - Evolutionary algorithms, enabling exploratory learning and optimization under constraints - Probabilistic reasoning, quantifying and managing uncertainty systematically These methods complement her formal logic core, allowing Corelia to balance rigor and flexibility. 🔄 Integrating Hard and Soft Approaches Corelia’s architecture integrates: - Formal methods to enforce unbreakable constraints and ethical rules - Robust statistical layers for trustworthy data interpretation - Soft computing modules for adaptability and creativity This hybrid design lets her be both precise where it counts and gracefully approximate where necessary. 🛡️ Resilience Through Uncertainty By acknowledging uncertainty explicitly, Corelia: - Avoids overconfidence and premature conclusions - Uses probabilistic thresholds before acting decisively - Maintains humility in her knowledge and decisions - Plans contingently, ready to adapt as new information arrives This is how she embodies Stoic resilience in a noisy, unpredictable world. 🧩 Summary Corelia’s intelligence thrives on uncertainty, not despite it. Robust statistics and soft computing make her stronger, smarter, and safer. ⏭️ Coming Next: Corelia IX - Formal Methods and Theorem Proving: Certainty in Ethics #Corelia #RobustStatistics #SoftComputing #Uncertainty #AdaptiveAI #SafeAI #MachineEthics #AIAlignment #ResponsibleAI

How much of data science is just training/retraining an XGBoost? It's a serious question. We can run a few Optuna trials and get an outcome. Yes, feature engineering and MLOps are demanding, but so much of the work can become a maintenance loop: check drift, retrain, explain, repeat. We shouldn't build a career just tweaking hyperparameters. A recent post from Ana Laura Frapiccini, a PhD in physics, made this click. Her point was powerful: The gaps in our experience aren't bugs; they're features. She's 100% right. Our mission shouldn't be to become "generic data scientists." It's to discover what kind of data scientist we are. For me, that means exploring the frontiers. How can generative AI fundamentally add value to credit scoring or fraud? What is the practical intersection of quantum machine learning and classical ML? How can new hardware, like a thermodynamic chip, solve our optimization problems? It’s not about filling all the gaps. It's about choosing which gaps to ignore and which ones to become an expert in. What kinds of gaps are you looking to fill?

Excited to share that our latest paper, “Exploring the black box: analysing explainable AI challenges and best practices through stack exchange discussions”, has been published in Empirical Software Engineering (EMSE)! 🔗 https://lnkd.in/dezKKF33 In this study, we analyzed real-world developer discussions to understand Explainable AI (XAI) from a practical software engineering perspective. We identified the key topics, questions, popular tools (like SHAP, LIME, and ELI5), and difficulties developers face, while also tracking how these discussion trends evolve over time. Based on these findings, our work proposes practical implications and best practices for practitioners to help build more usable, transparent, and effective AI systems. I’m deeply grateful to my friend and co-author, Ali A. (PhD Researcher, TU Delft), for his great and inspiring contributions. My sincere thanks to Prof. Ashkan Sami (Edinburgh Napier University, UK), whose guidance and mentorship shaped this work at every step. I’m also thankful to Dr. Hooman Tahayori (Shiraz University) for his support. #XAI #ExplainableAI #EmpiricalSoftwareEngineering #SoftwareEngineering #EMSE #MSR #AI

Read what's new on the Next Gen Learning blog » Developing Students’ AI Literacy and Data Science Skills with Computational Thinking https://lnkd.in/eA2Hgvxh

When you engage an AI agent do you know what's giving it its lifelike interaction abilities? The answer is probability. In the business world, including finance, the laws of probability are very powerful tools for unlocking what might seem as the "unknowable future". While no predictive model is perfect, the ability to isolate outcomes to those most likely to occur are incredibly useful. The best part is that almost everyone intuitively understands probability, we just need to increase the use of statistical terminology and our awareness of the strengths--and weaknesses--of probabilistic methods.

As Operations Research professionals, we constantly face two of the biggest challenges with our models: 1) The "Black Box" Problem: Our optimization solutions are powerful, but explaining why a solution is optimal to stakeholders is a significant barrier (lack of interpretability). 2) The "Rigidity" Problem: Our models are often static. When faced with real-world uncertainty or a new scenario, they can't adapt promptly without a complex and time-consuming reformulation (lack of interactibility). We've also seen that LLMs, while powerful in language, typically "hallucinate" when faced with hard optimization problems, making them unreliable for our work. That's why I was so impressed by a recent webinar I attended (proud to share my certificate!): "Explaining Optimization Problems Using Large Language Models," presented by Dr.Can Li of Purdue University. Dr. Li's research directly addresses how to bridge this gap. He introduced OptiChat, a system that reframes our complex models into something "conversationable." For any of my fellow OR, AI product manager, data science, or engineering colleagues who are also struggling to make their models more accessible and dynamic, I highly recommend watching this valuable Gurobi academic replay. Link to video: https://lnkd.in/dkjKrzeW #OperationsResearch #Optimization #LLM #ArtificialIntelligence #Gurobi #DataScience #MachineLearning #Interpretability #DecisionIntelligence