Steve Smith

We at Northwestern University Computer Science Department were delighted to host shafi goldwasser, winner of the ACM, Association for Computing Machinery Turing Award, and Director of the Simons Institute for the Theory of Computing here on Friday May 10. She also shared with us the sad story that Jim Simons, mathematician and founder of Renaissance Technologies LLC passed away yesterday at the age of 86. Rest in peace Jim. Read more about Jim Simons here - https://lnkd.in/gPXrM6n8 #simons #acm Here talk was about "Trust, Backdoor Vulnerabilities and Possible Mitigations". Abstract: Cryptographic tools and models enable to trust the use of technology platforms controlled by worst case computationally bounded adversaries. In this talk I will use cryptographic modeling and tools to view trust dilemmas in various phases of the machine learning pipelines. We will touch on privacy in the training stage, verification of properties of machine learning models, and the possibility of achieving robustness in presence of backdoors.

81

If you are attending ACM SIGMETRICS 2024 this week, don't miss the Sustainable Computing session (aka the "UMass session"), where UMass students will present all three papers in the session! Adam Lechowicz will present "The Online Pause and Resume Problem: Optimal Algorithms and an Application to Carbon-Aware Load Shifting" Read it here: https://lnkd.in/dBQ4C93F Walid Hanafy will present "CarbonScaler: Leveraging Cloud Workload Elasticity for Optimizing Carbon-Efficiency." Read it here: https://lnkd.in/dRzFCNux Adam Lechowicz will also present "Online Conversion with Switching Costs: Robust and Learning-Augmented Algorithms" Read it here: https://lnkd.in/dg3c2p66 Thanks to all co-authors: Noman Bashir David Irwin Adam Wierman Mohammad Hajiesmaili Qianlin Liang #sustainability #cloudcomputing

68

2 Comments

Big congrats to my colleague, Assistant Professor Hannah Kerner. She and her team received the Best Paper Award for their work, "Sat2Cap: Mapping Fine-Grained Textual Descriptions from Satellite Images," at the EarthVision workshop at the CVPR 2024 conference. Designed to provide a weakly supervised approach for creating maps using free-form textual descriptions, the team’s research offers experimental results and demonstrates models that successfully capture ground-level concepts, allowing large-scale mapping of fine-grained textual queries. This approach does not require any text-labeled data, making the training easily scalable. Great work! https://lnkd.in/gKc3shmU School of Computing and Augmented Intelligence

44

Daily trivia 11: AI/ML and Data Science are popular topics these days, but our Systems Science program isn't just about AI/ML/DS. Students will also acquire the ability of "mechanistic modeling" of complex systems, a key skill for complex decision-making. https://lnkd.in/dYEJr8CM

7

This is a very interesting article and a very important conversation. It brings up topics tangential to things that have been on my mind lately: the future of academia and the future of applied science jobs.

2

Congratulations, Matthew DeLorenzo, on getting your first paper accepted and open-sourcing the code. While LLMs are increasingly used for code generation across various domains, there's a notable gap: few are evaluating the creativity of these codes. If you wish to assess the "creativity" of LLMs in generating Verilog code, here is a (incomplete) set of metrics. https://lnkd.in/gR6bjbNZ Here is the pre-print of the paper accepted at LAD. https://lnkd.in/gdbXAgSp

17

Let’s talk LLM research.

2

AI is here and here to stay. For sure something like that has been said about calculators decades ago, but maybe this talking, scaffolding AI will finally get us to seriously rethink and redesign (Math) education. See how smooth the student-AI tutoring conversation went? That is thanks to Imran’s good Math communication ability — just one of what students need to thrive and make the most of their artificial intelligence-infused reality. The world will no longer be as it is today. If formal education doesn’t keep up, no matter how much we romanticize the human touch of teaching, this generation (Z, Alpha, and those to come) might grow up choosing the consistently competent, and in all fairness, even expressively caring, AI for the academic learning of their children. P.S. GPT-4o’s scaffolding can be improved but wow, that’s a pretty good start. P.P.S. Maybe let’s move forward with… AI is not an enemy. 😊 https://lnkd.in/gR_46V_P #chatgpt #gpt4 #omni #AI #education

1

This just got published at CHI '24: Framing choice options as a creative process increases perceived customization. This work (from Prof. Alice Moon) has been some of my favorite research to talk about over the years! https://lnkd.in/e8DnAYdi

53

📣 New publication out 📣 A chapter I have written together with my former colleague Henning Mayer just got published in the Oxford University Press Handbook of the Sociology of Machine Learning edited by Christian Borch and Juan Pablo Pardo-Guerra. In it, we argue that contemporary forms of #machinelearning challenge core assumptions of #socialtheory. We illustrate how a #sociology of #interfaces can tackle some of these issues by studying #human #machine interaction at, through, and across the interface.

35

4 Comments

Brain Power & Heart Power: A Deep Dive into Learning Design & Technology Two brilliant minds, two incredible mentors: Professor Tutaleni I. Asino and Professor Marti Louw, from Carnegie Mellon's Learning Science team, have left me feeling both energized and humbled! 🧠💡 I walked into this deep dive with a thirst for knowledge, and I left with my mind buzzing. Learning about adaptive learning systems and AI integration in education was mind-blowing. The way Professor i.Asino explained it, it's like giving every student their own personal tutor – how cool is that?! 🤯 Professor Louw's insights about the impact of technology on learning, and how we can empower students to become their own learning guides (metacognition!), were incredibly inspiring. Thinking about how portfolios can be used to help students reflect on their learning journey, it's truly powerful stuff. This experience has given me a fresh perspective on instructional design. I'm super excited to dive deeper into user-centered design principles and learning analytics. It's time to step up my game and make a real impact in this field! Huge thanks to the Learning Science team for this amazing opportunity! I feel so lucky to be part of this incredible community. ✨ #learningdesign #instructionaldesign #CarnegieMellon #mentorship #inspired #challenge #futureoflearning

17

Following Jee's meesage, please let me share a new related work jointly done with Tomi Kinnunen, Kong Aik Lee, Paul-Gauthier Noé, and Junichi Yamagishi. 📜 Title Revisiting and Improving Scoring Fusion for Spoofing-aware Speaker Verification Using Compositional Data Analysis (arxiv link https://lnkd.in/g568pKXB) ⚒ Use it now !? The proposed fusion method can be used in the ongoing ASVspoof 5 challenge, Track 2 (https://www.asvspoof.org/). 🔗 link to the paper: https://lnkd.in/g568pKXB 🔗 a Jupyter notebook: https://lnkd.in/gW8bPdfJ 🔗 code of SASV system: https://lnkd.in/gPyh6NPh There will be a package on ASVspoof5 github site as well https://lnkd.in/gU6XUhhG (currently in private) 🔖 Background:  The paper done with Jee showed how ASV systems trained on bona fide data (human speech data) could defend themselves from spoofing attacks to some extent. I understand this potential is consistent with the expectation of many researchers, but currently, it may still be helpful to have spoofing-robust ASV systems, either built upon the combination of a spoofing countermeasure (CM) and an ASV system or trained using the bona fide and spoofing data in an end-to-end manner. 🆕 Key question: how should we fuse CM and ASV scores? Our new work revisits this question and produces findings from ** a decision theory point of view **. We describe a non-linear fusion of ASV and CM log-likelihood-ratios-like quantities, which minimizes the conditional risk in decision making for the task. 📚 The new paper also links to many pieces of work by others, from which I learned a lot. I will write more in the threads below. My understanding on related topics is still shallow. Looking forward to everyones' comments!

31

8 Comments

"Wiley, a 217-year-old publisher based in Hoboken, N.J., ... Tuesday will announce that it is closing 19 journals, some of which were infected by large-scale research fraud" according to an exclusive report by The Wall Street Journal. For years, some of us have been discussing systemic issues plaguing the "scientific" literature, usually to be ignored, even belittled, by publishers, editors, and peers. Well, here we are then: "In the past two years, Wiley has retracted more than 11,300 papers that appeared compromised, according to a spokesperson, and closed four journals." Maybe this detailed reporting from a major news source will attract more attention from people outside narrow technical fields. The stakes couldn't be higher. I've lost track of all the wild articles shared on LinkedIn about doing fake science with, for example, ChatGPT. Career columns have encouraged early career researchers to engage in scientific misconduct and have a chatbot write their grant or paper (I'm not exaggerating: https://lnkd.in/gZEbGyfx). Rest assured, such scientific dishonesty will catch up to those "scientists." But, for now, the battle is hardly won, as the Journal reports: "While publishers are fighting back with technology, paper mills are using the same kind of tools to stay ahead. “Generative AI has just handed them a winning lottery ticket,” Eggleton of IOP Publishing said. “They can do it really cheap, at scale, and the detection methods are not where we need them to be. I can only see that challenge increasing.”" https://lnkd.in/guyyafBX

27

3 Comments

While I think digital twins are a good idea for early research and development, it is CRITICAL to remember that for any system with embedded artificial intelligence, digital twins will not be sufficient for testing systems expected to operate in the real world. Such systems should NEVER be allowed to be certified for operations based only on simulated data.

75

26 Comments

Fascinating research. “For learning to be useful, it has to be remembered,” added Steve Chase, professor of biomedical engineering and the Neuroscience Institute at Carnegie Mellon University and one of the three leads on the project, along with Byron Yu of Carnegie Mellon and Aaron Batista of the University of Pittsburgh. “The process of how you take a learned experience and lay it down in neural activity, so it can be recalled again when you need it, is a fascinating process. Our understanding of it is still in its infancy. The hope is that by looking at how these learned acts are laid down for motor tasks, it will help us know how they’re laid down for other tasks as well. And, ultimately, once we’re able to do all of this, we can try to understand what all the factors that help us remember are, to really track the memory process itself.” https://lnkd.in/gGqdw5-D #learning #memory

8

3 Comments

A while back, I posted a paper here regarding the use of natural language processing (NLP) for a crosswalk study (Common Core State Standards, and NAEP item specifications). However, I found that reading the paper on a laptop in the posted format was not convenient. As a result, I submitted it to arxiv: https://lnkd.in/dgByjKQD The paper is available in pdf as well as HTML. A few notes on manuscript preparation: I prepared the paper in latex using the free service at Overleaf, which provides an arxiv template as well as the facility to submit directly to arxiv. I found it was easy to manage the revision process including several rounds of correcting substantive errors and grammatical glitches. A paper sometimes takes on a life of its own and publication requires a decision for a hard stop. It's a blessing and a curse. Here is the abstract: "Natural language processing (NLP) is rapidly developing for applications in educational assessment. In this paper, I describe an NLP-based procedure that can be used to support subject matter experts in establishing a crosswalk between item specifications and content standards. This paper extends recent work by proposing and demonstrating the use of multivariate similarity based on embedding vectors for sentences or texts. In particular, a hybrid regression procedure is demonstrated for establishing the match of each content standard to multiple item specifications. The procedure is used to evaluate the match of the Common Core State Standards (CCSS) for mathematics at grade 4 to the corresponding item specifications for the 2026 National Assessment of Educational Progress (NAEP)."

14

RELEASED TODAY! My book, Mixed Methods for Psychological Methods, is finally out! It presents theory and methods on how we can bring qualitative perspectives directly into measurement with applications for #CritQuant and #SituatedAssessment. https://lnkd.in/gW-TsHJQ

72

8 Comments

Daily reminder that simple semantic detection is not the solution to AI "plagiarism" or detection. The copy/pasters, cyborgs, and centaurs will adapt; the AI will adapt. As educators (and humans) we need to rethink what the future is going to look like with this new intelligence. (Credit to Ethan Mollick for the cyborg and centaur terms. Excellent book on the future AI collaboration, I'd highly recommend it!) Note: The point of this post is to demonstrate the futility and potential risk of using simple semantic detection, not an endorsement to unethically use AI. Chat GPT Cliches: https://lnkd.in/gRmzB5eF

8

1 Comment