Implement publications section

pyproject.toml

@@ -5,6 +5,7 @@ description = "SSL (Software and Systems Laboratory) website "
 readme = "README.md"
 requires-python = ">=3.14"
 dependencies = [
+    "ablog>=0.11.13",
     "sphinx>=8.2.3",
     "sphinx-autobuild>=2025.8.25",
     "sphinx-book-theme>=1.1.4",

src/conf.py

@@ -16,6 +16,7 @@
 # https://www.sphinx-doc.org/en/master/usage/configuration.html#general-configuration
 
 extensions = [
+    "ablog",
     "sphinx.ext.autosectionlabel",
     "sphinx.ext.intersphinx",
     "sphinx.ext.todo",

src/index.rst

@@ -366,6 +366,6 @@ consult Loyola University Chicago's `official admissions resources <https://gpem
    :caption: Pages
    :hidden:
 
+   publications/index.rst
    members
    projects
-   publications

src/publications/agentic-tutor-hpc-sc25-eduhpc2025.rst

@@ -0,0 +1,120 @@
+:blogpost: true
+:date: November 29, 2025
+:category: Blog Post
+:tags: Artificial Intelligence, 11-29-2025
+:nocomments:
+
+======================================================================
+Advancing HPC Education with an Agentic Tutoring System (EduHPC 2025)
+======================================================================
+
+This post highlights a recent **EduHPC 2025** paper `doi:10.1145/3731599.3767386 <https://dl.acm.org/doi/10.1145/3731599.3767386>`__ led by my PhD student **Erik Pautsch** and co-supervised by me and  **Silvio Rizzi** at Argonne National Laboratory.  
+
+In this work, we introduce an **agentic tutoring system** that supports instructors in planning, teaching, and assessing high-performance computing (HPC) content.  
+Our goal is to make HPC instruction more accessible, scalable, and sustainable as it is presently lacking in all three dimensions.
+We think AI can help.
+
+Abstract / Summary
+------------------
+
+This summer we explored how an agentic tutoring system could lower barriers to HPC and parallel/distributed computing (PDC) education.  
+Our EduHPC 2025 paper presents a system that assists instructors with:
+
+- lesson planning  
+- generation of teaching materials  
+- adaptive explanations based on learner background  
+- structured assessment and feedback  
+
+Our initial deployment at Argonne demonstrates that we can prepare lessons more efficiently, improve accessibility of HPC concepts, and streamline evaluation.  
+We believe this represents a meaningful step toward **democratizing HPC education**.
+
+Background: Why HPC Needs Better Teaching Tools
+-----------------------------------------------
+
+HPC and PDC skills are increasingly essential across scientific computing, engineering, modeling and simulation, and data-intensive research.  
+
+Yet teaching HPC remains challenging for many reasons:
+
+- Many institutions lack faculty with substantial HPC experience.  
+- Preparing examples for MPI, OpenMP, GPUs, profiling, and performance studies is labor-intensive.  
+- Existing curricula are fragmented and often tied to individual experts.  
+- Adoption of HPC coursework is still uneven, especially at teaching-focused institutions.
+
+Our work asks whether **intelligent tutoring systems can meaningfully support instructors** -- reducing preparation time, lowering expertise requirements, and expanding access.
+
+Key Contributions
+-----------------
+
+Our paper makes several core contributions:
+
+1. **We created an agentic HPC tutoring system** that helps with lesson planning, content generation, instructional guidance, and assessment.  
+2. **We designed the system explicitly for instructors of varying experience levels**, not only HPC specialists.  
+3. **We implemented adaptive content generation**, allowing material to adjust for novice or advanced learners.  
+4. **We integrated assessment capabilities** that support consistent evaluation of student work.  
+5. **We conducted an evaluation at Argonne**, demonstrating feasibility under realistic instructional conditions.
+
+Initial Results
+---------------
+
+Our early results show:
+
+- Lesson plans and teaching units could be generated significantly faster than preparing materials manually.  
+- The generated instructional materials—slides, examples, exercises—were high enough quality for direct classroom use.  
+- Adaptivity helped us reach learners with different backgrounds effectively.  
+- The system performed well in a realistic (but small) instructional environment.
+
+These findings suggest that intelligent systems can **augment instructors**, helping expand HPC education capacity.
+
+Key Take-Aways
+--------------
+
+- Agentic systems can broaden access to HPC education across many types of institutions.  
+- Automating planning and assessment reduces barriers to offering HPC courses.  
+- Adaptivity improves inclusion by addressing diverse learner backgrounds.  
+- The Argonne deployment on leadership-class systems (e.g. Polaris) demonstrated real-world practicality.  
+
+Future Work
+-----------
+
+We plan to extend this work by:
+
+- Deploying the system across additional academic settings  
+- Conducting longitudinal studies over full terms  
+- Expanding HPC topic coverage (MPI, OpenMP, GPUs, profiling, debugging, energy-aware computing)  
+- Adding automated grading and adaptive exercises  
+- Integrating with learning management systems  
+- Supporting community-driven lesson templates and shared teaching modules  
+
+Access
+----------------------
+
+.. note:: You should be able to downlaod this paper with or without an ACM Digital Library subscription.
+   Please contact me if you cannot do so.
+
+- `DOI 10.1145/3731599.3767386 <https://dl.acm.org/doi/full/10.1145/3731599.3767386>`__
+- `PDF <https://dl.acm.org/doi/pdf/10.1145/3731599.3767386>`__
+
+Citation
+----------------------
+
+Erik Pautsch, Mengjiao Han, Joseph A. Insley, Janet Knowles, Victor A. Mateevitsi, Silvio Rizzi, and George K. Thiruvathukal. 2025. *An Interactive Agentic HPC Tutor for Lesson Planning, Teaching, and Assessment*. In Proceedings of the SC '25 Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC Workshops '25). Association for Computing Machinery, New York, NY, USA, 367–375. https://doi.org/10.1145/3731599.3767386
+
+.. code-block:: bibtex
+
+   @inproceedings{10.1145/3731599.3767386,
+      author = {Pautsch, Erik and Han, Mengjiao and Insley, Joseph A. and Knowles, Janet and Mateevitsi, Victor A. and Rizzi, Silvio and Thiruvathukal, George K.},
+      title = {An Interactive Agentic HPC Tutor for Lesson Planning, Teaching, and Assessment},
+      year = {2025},
+      isbn = {9798400718717},
+      publisher = {Association for Computing Machinery},
+      address = {New York, NY, USA},
+      url = {https://doi.org/10.1145/3731599.3767386},
+      doi = {10.1145/3731599.3767386},
+      abstract = {High-performance computing (HPC) education is at an inflection point, driven by agentic systems and “prompt-engineering” as a form of programming. We describe an interactive tutor built from autonomous LLM-based agents, each with a narrow role: planning lessons, explaining concepts, scaffolding code, and executing runs. Using open-source toolkits and locally hosted models on leadership-class supercomputers, the tutor lets educators generate and refine parallel-programming examples in real time without external APIs or subscription fees. Complex workflows are composed through structured prompts rather than traditional source code, while per-agent history summarization prevents context-window overflow and enables self-correcting code generation. Requiring no proprietary services, the platform is immediately deployable in institutional HPC environments and scales from single-user sessions to classroom labs. Beyond a teaching aid, it illustrates how prompt-driven, multi-agent software can deliver dynamic, personalized, and extensible learning experiences across technical domains.},
+      booktitle = {Proceedings of the SC '25 Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis},
+      pages = {367–375},
+      numpages = {9},
+      keywords = {high-performance computing, multi-agent systems, large language models, education, CUDA, SYCL},
+      location = {St. Louis, MO, USA},
+      series = {SC Workshops '25}
+      }

src/publications/ai-in-hiring-fairness-or-just-automated-bias.rst

@@ -0,0 +1,64 @@
+:blogpost: true
+:date: November 18, 2024
+:category: Blog Post
+:tags: Artificial Intelligence, 11-18-2024
+:nocomments:
+
+AI in Hiring: Fairness or Just Automated Bias?
+==============================================
+
+Artificial intelligence has become increasingly embedded in modern hiring systems.
+From résumé screening to candidate scoring, automated tools promise efficiency, objectivity, and scale.
+Yet these promises often obscure important risks: when AI models inherit biased historical data, they can reinforce or even amplify inequities in hiring.
+
+Many hiring datasets reflect systemic social, cultural, and economic disparities.
+If an organization’s historical hiring patterns favored one demographic group—intentionally or not—an AI system trained on that data is likely to replicate those preferences.
+Under the guise of neutrality, the model may recommend “more of the same,” reducing diversity and overlooking equally qualified candidates.
+
+This raises a crucial question:
+**Are AI hiring tools fair, or are they simply automating existing forms of bias?**
+
+Important concerns
+------------------
+
+* **Historical bias baked into training data**  
+  AI systems inherit the limitations and inequities of the datasets used to train them.
+
+* **Opacity and lack of accountability**  
+  Candidates often cannot understand, challenge, or appeal algorithmic decisions.
+
+* **Risk of reinforcing homogeneity**  
+  Automated systems may unintentionally filter out qualified applicants whose backgrounds differ from past hires.
+
+* **Regulatory and legal implications**  
+  As governments introduce stricter rules for automated hiring systems, organizations must ensure transparent and fair processes.
+
+Moving forward responsibly
+---------------------------
+
+To ensure responsible AI use in hiring, organizations must:
+
+* audit models regularly for disparate impact,  
+* adopt transparent scoring criteria,  
+* maintain meaningful human oversight, and  
+* prioritize fairness and inclusivity in both design and deployment.
+
+AI can assist in hiring, but it must never replace critical human judgment—especially when people’s careers and livelihoods are at stake.
+
+Citation
+--------
+
+Theresa Fister and George K. Thiruvathukal, "Artificial Intelligence Employment Interviews: Examining Limitations, Biases, and Perceptions," in Computer, volume 57, number 10, pages 76-81, October 2024, https://doi.org/10.1109/MC.2024.3404669.
+
+.. code-block:: bibtex
+
+   @article{10687332,
+     author={Fister, Theresa and Thiruvathukal, George K.},
+     journal={Computer}, 
+     title={Artificial Intelligence Employment Interviews: Examining Limitations, Biases, and Perceptions}, 
+     year={2024},
+     volume={57},
+     number={10},
+     pages={76-81},
+     keywords={Artificial intelligence;Employment;Training;Quality assessment;Software quality;Business},
+     doi={10.1109/MC.2024.3404669}}

src/publications/index.rst

@@ -0,0 +1,9 @@
+################
+ Blog Posts
+################
+
+.. postlist::
+ :date: %A, %B %d, %Y
+ :format: {title}
+ :excerpts:
+ :expand: Read more ...

src/publications/model-naming-emse-2025.rst

@@ -0,0 +1,84 @@
+:blogpost: true
+:date: December 27, 2025
+:category: Blog Post
+:tags: Software Engineering, 12-27-2025
+:nocomments:
+
+PTM Naming: Why “What’s in a Name” Actually Matters for AI Reuse
+==================================================================
+
+I’m thrilled to share some recent work led by **Wenxin Jiang**, a PhD
+student at Purdue University. Wenxin is supervised by **James C.
+Davis**, and I have had the pleasure of serving as a key external
+supervisor and PhD committee member on this project as part of my
+ongoing collaboration with Dr. Davis.
+This research was recently accepted for publication in `Journal of Empirical Software Engineering <https://link.springer.com/journal/10664>`__ and it tackles a problem that anyone working in
+AI has likely grumbled about: how we name our models.
+
+If you’ve spent any time on registries like **Hugging Face**, you know
+the drill: you need a model for a specific task, you search for
+something like “BERT,” and you’re suddenly staring at a wall of names
+like ``bert-base-uncased``. In traditional software, we’re used to
+short, catchy names like ``requests`` or ``chalk``. But in the world of
+deep learning, the name is often the only “spec sheet” a developer has
+to infer critical details about architecture, size, and datasets.
+
+The Wild West of Model Names
+-----------------------------
+
+Wenxin’s study—which included a survey of 108 Hugging Face users and a mining study of over 14,000 model packages—confirms that Pre-Trained Model (PTM) naming is a completely different from traditional software conventions.
+While a traditional package name tells you “what it does,” a PTM name tries to pack in “how it works,” “how big it is,” and “what it was trained on” all at once.
+
+The problem is that we are currently in a bit of a “Wild West” environment when it comes to PTM naming.
+Our research found a massive mismatch between what engineers *want* in a name and what they actually *get*.
+For instance, most developers prefer names that reflect both the implementation and the intended task, but in practice, names tend to stick almost exclusively to technical architectural units.
+Even worse, these names are often inconsistent; we’ve seen cases where a model’s identifier doesn’t match its internal metadata, leading to significant debugging effort for users.
+
+DARA: An Automated Consistency Detector
+-----------------------------------------
+
+To help clean up this ecosystem, our team developed **DARA (DNN ARchitecture Assessment)**.
+The idea is simple but powerful: can we look at the actual “guts” of a model—its layers and connections—and predict what its name *should* be?.
+
+It turns out, we can. By treating the model’s computational graph as an “abstract architecture” (APTM), DARA can detect inconsistencies with surprising accuracy.
+Our results show that architectural information alone is sufficient to achieve an accuracy of **99%** in identifying the correct ``model_type``.
+
+This isn’t just an academic exercise.
+We envision tools like DARA being used as “linters” during the model upload process.
+Imagine uploading a model to a registry and getting an instant alert: “Hey, you named this model ‘nllb,’ but its architecture metadata says ‘m2m.’
+Want to double-check that?”.
+
+Why This Matters for the Supply Chain
+-----------------------------------------
+
+Beyond just making it easier to find the right model, this is a security issue.
+As PTMs become core components of our software supply chain, wee have to worry about things like **typosquatting** and **architectural backdoors**.
+If an attacker can hide a malicious model under a familiar-looking name, and we don’t have automated ways to verify that the name matches the content, the entire ecosystem is at risk.
+
+Wenxin’s work provides the first empirical foundation for standardizing how we talk about and name these models.
+It’s a major step toward making AI reuse as reliable and transparent as traditional software engineering.
+
+If you’re interested in the technical nitty-gritty or want to try out
+the DARA tool yourself, the paper and the code are available at 
+`GitHub PurdueDualityLab/PTM-Naming <https://github.com/PurdueDualityLab/PTM-Naming>`__.
+
+Citation
+~~~~~~~~
+
+Jiang, Wenxin, Mingyu Kim, Chingwo Cheung, Heesoo Kim, George K. Thiruvathukal, and James C. Davis. *"I See Models Being a Whole Other Thing": An Empirical Study of Pre-trained Model Naming Conventions and a Tool for Enhancing Naming Consistency*. Empirical Software Engineering, vol. 30, no. 6, 2025, p. 155. Springer Link, https://doi.org/10.1007/s10664-025-10711-4.
+
+.. code-block:: bibtex
+
+   @article{Jiang2025,
+     author    = {Jiang, Wenxin and Kim, Mingyu and Cheung, Chingwo and Kim, Heesoo and Thiruvathukal, George K. and Davis, James C.},
+     title     = {``I see models being a whole other thing'': An empirical study of pre-trained model naming conventions and a tool for enhancing naming consistency},
+     journal   = {Empirical Software Engineering},
+     year      = {2025},
+     volume    = {30},
+     number    = {6},
+     pages     = {155},
+     month     = {aug},
+     doi       = {10.1007/s10664-025-10711-4},
+     url       = {https://doi.org/10.1007/s10664-025-10711-4},
+     issn      = {1573-7616}
+   }