Francesca Rossi

Jorge Murillo presented our paper titled "CDN-Shifter: Leveraging Spatial Workload Shifting to Decarbonize Content Delivery Networks" at the ACM Symposium on Cloud Computing (ACM SoCC) earlier this week. The paper presents a spatial workload scheduling algorithm called CDN-Shifter for greening CDNs. Read it here: https://lnkd.in/eSxCU-Qh Walid Hanafy David Irwin Ramesh Sitaraman #cloudcomputing #decarbornization

46

Our latest IEEE TTS paper is by Staff Research Scientist Kevin McKee at Google DeepMind, who is focused on developing #cooperative and #inclusiveAI. If you are engaged in the secondary use of #data (e.g., #health/ #financial/ #taxpayer/ #education), or are dealing with data that has been provided by a #humansubject, and working in the #AI / #ML space, then this is a must-read article. ➡ https://lnkd.in/gFi6sniw The claim is made by the author that fewer than 1 out of 4 #AAAI and #NeurIPS papers confirm independent ethical review, the collection of informed consent, or participant compensation. We might then pose the question: what's gone wrong in the process of publication at these conferences, if anything at all? From where does the problem stem? What can be done to curb the rampant use of primary data without the consent of the participant? Excerpt: In recent years, research involving human participants has been critical to advances in AI and ML, particularly in the areas of conversational, human-compatible, and cooperative AI. For example, roughly 9% of publications at recent AAAI and NeurIPS conferences indicate the collection of original human data. Yet AI and ML researchers lack guidelines for ethical research practices with human participants. Fewer than one out of every four of these AAAI and NeurIPS papers confirm independent ethical review, the collection of informed consent, or participant compensation. This paper aims to bridge this gap by examining the normative similarities and differences between AI research and related fields that involve human participants. Though psychology, human-computer interaction, and other adjacent fields offer historic lessons and helpful insights, AI research presents several distinct considerations—namely, participatory design, crowdsourced dataset development, and an expansive role of corporations—that necessitate a contextual ethics framework. Citation: K. R. McKee, "Human Participants in AI Research: Ethics and Transparency in Practice," in IEEE Transactions on Technology and Society, vol. 5, no. 3, pp. 279-288, Sept. 2024, doi: 10.1109/TTS.2024.3446183. Keywords: #Ethics #Psychology #Guidelines #Companies Rob Nicholls George Roussos Rafael A. Calvo Jason Sargent Ehsan Nabavi Suchit Ahuja Dr Greg Adamson Ruth Lewis Peter Lewis Peter Asaro Evan Selinger Julia Powles Michael Zimmer Constance de Saint Laurent Edgar Whitley Professor Carolyn McGregor AM 💡 Rebecca Herold Deborah Lupton Sonia Livingstone Carole McCartney Susan Halford FAcSS, FRSA Michael King Grace McCarthy Lucy Batley Dr Genevieve Smith-Nunes Bozenna Pasik-Duncan Lyria Bennett Moses Dilan Thampapillai Andrew Whelan Savvas Papagiannidis Dinara Davlembayeva Ilias Pappas Marijn Janssen Efpraxia Zamani Anastasia Griva Patrick Mikalef

96

9 Comments

Thrilled to announce the publication of our latest research paper titled "An empirical evaluation of tinyML architectures for Class-Incremental Continual Learning" , which was accepted at the IEEE Annual Conference on Pervasive Computing and Communications Workshops (PerCom) held in Biarritz, France. Check out the publication at: https://lnkd.in/d9WpM_qe Huge thanks to my brilliant co-authors Matteo Tremonti, Davide Dalle Pezze, Francesco Paissan, and Elisabetta Farella for their invaluable contributions! In this paper, we delve into the realm of tinyML architectures and their potential for solving the challenges of Class-Incremental Continual Learning. Our study focuses on evaluating the performance of PhiNet models on CORe50 and CIFAR-10 benchmarks, showcasing their superiority over MobileNet architectures in terms of both accuracy and computational efficiency. PhiNet not only achieved a 4.47% higher accuracy on CIFAR-10 but also demonstrated remarkable computational efficiency, utilizing only 0.012% of the computation required by MobileNet. These findings underscore the promising prospects of deploying PhiNet models in real-world edge device applications. Looking forward to further advancements in the field of tinyML and continual learning. AMCO Lab - UniPD, Fondazione Bruno Kessler - FBK, Dip. di Ingegneria dell'Informazione (DEI) - Dept. Information Engineering / UniPD #AI #TinyML #ContinualLearning #Research #IEEEPerCom #EdgeComputing

44

💥💥💥 Duo-LLM: A Framework for Studying Adaptive Computation in Large Language Models Keivan Alizadeh, Iman Mirzadeh, Hooman Shahrokhi, Dmitry Belenko, Frank Sun, Minsik Cho, Mohammad Hossein Sekhavat, Moin Nabi, Mehrdad Farajtabar Abstract Large Language Models (LLMs) typically generate outputs token by token using a fixed compute budget, leading to inefficient resource utilization. To address this shortcoming, recent advancements in mixture of expert (MoE) models, speculative decoding, and early exit strategies leverage the insight that computational demands can vary significantly based on the complexity and nature of the input. However, identifying optimal routing patterns for dynamic execution remains an open challenge, limiting the full potential of these adaptive methods. To address this need, we study adaptive computation in LLMs more systematically. We propose a novel framework that integrates smaller auxiliary modules within each Feed-Forward Network layer of the LLM. This design enables dynamic routing of tokens based on task complexity: tokens can be processed by either the small or big modules at each layer, or even bypass certain layers entirely. This allows us to introduce a novel notion of a token's difficulty, defined by its potential to benefit from additional computational resources. Importantly, by employing oracles to identify optimal patterns of adaptive computations, we gain valuable insights into the internal workings of LLMs and the routing processes in a simplified heterogeneous MoE setup. We show that trained routers operate differently from oracles and often yield suboptimal solutions. Notably, activating a large module in just one layer outperforms models that use large modules across all layers, underscoring the gap between practical implementations of routing in MoE models and theoretical optima for adaptive computation. 👉 https://lnkd.in/d7-GhpCz #machinelearning

1

Interesting opportunity: inviting paper submissions on "Quantum sensors and quantum computing applications in IoT and Smart Cities" to be presented in the 3rd Workshop on Quantum in Consumer Technology at IEEE Quantum Week 2024.

13

Applied Affective Computing https://bit.ly/3naKZQw provides State-of-the-art and emerging themes in affective computing, including a comprehensive review of the existing approaches to affective computing systems and social signal processing. Authors: Leimin Tian, Sharon Oviatt, Michal Muszynski, Brent Chamberlain, & Jennifer Healey. #AI #AffectiveComputing #MachineLearning #SocialRobots #BuiltEnvironments #Multimodal #DataCollection #ReinforcementLearning #Synthesizing #Natural and #Believable #EmotionalExpressions ACM, Association for Computing Machinery

1

📣 Kiroong Choe is presenting our new IEEE TVCG paper, Enhancing Data Literacy On-demand: LLMs as Guides for Novices in Chart Interpretation, at the "Look, Learn, Language Models" session this Friday: https://lnkd.in/eygdA_jt 📝 Author List: Kiroong Choe, Chaerin Lee, Soohyun Lee, Jiwon Song, Aeri Cho, Nam Wook Kim, Jinwook Seo 🎓 Kiroong (https://kiroong.github.io/), a Ph.D. student at Seoul National University advised by Prof. Jinwook Seo, spearheaded this work by leading a group of other students. I enjoy meeting with him because he always thinks critically about my input and offers different perspectives when needed, which in turn improves my own thinking. 🙂 📊 We compared three different methods of learning assistance for interpreting complex visualizations for novices: 1) using web search, 2) LLMs with text-based support, and 3) LLMs with both text and visual support. We developed a "visual prompting" system to support the third condition. 💡 An interesting finding is that, despite participants preferring the LLM-based approach, when we counted the number of insights, the web search performed the best, even though it was more time-consuming and challenging to use. 🤔 This might suggest that cognitive difficulty leads to more engagement and thus more learning. This would be an important consideration for future LLM-based learning systems. 📄 Please check out the full paper manuscript: https://lnkd.in/etfn6yGg

18

We invite research submissions on Quantum Computing and Reinforcement Learning (QCRL) at IEEE Quantum Week 2024 (Montreal, Sep. 15-20, 2024). https://2024.qcrl.io/ Scope and Topics: Topics relevant to this session include, but are not limited to: - Theoretical studies of quantum reinforcement learning algorithms - Challenges in quantum reinforcement learning in the NISQ era - Quantum reinforcement learning under the influence of quantum device noise QRL in the context of trustworthy ML (e.g., adversarial attacks, federated learning) - Quantum reinforcement learning for scientific discovery, commercial, and industrial applications - Benchmarking QRL algorithms - Reinforcement learning for quantum control, quantum error correction, and quantum error mitigation - Reinforcement learning for quantum architecture search, optimization, and program synthesis - Hybrid RL-QC systems design Important Deadlines: - Jul 10, 2024: Abstract Due Date - Jul 15, 2024: Paper Submission Due Date - Jul 28, 2024: Notification of Paper Acceptance - Aug 5, 2024: Camera-ready Papers Due #quantum #machinelearning #reinforcementlearning

30

After a year of engaging discussions, brainstorming sessions, workshops, and close collaborations between IBM Research and UIUC researchers within the IIDAI Institute, we are excited to present our Vision White Paper — Transforming the Hybrid Cloud for Emerging AI Workloads. We envision transforming hybrid cloud systems to meet the growing complexity of AI workloads through innovative, full-stack co-design approaches, emphasizing usability, manageability, affordability, adaptability, efficiency, and scalability. Collaborative efforts between academia and industry are central to this vision, driving advancements in foundation models for material design and climate solutions, scalable multimodal data processing, and enhanced physics-based AI emulators for applications like weather forecasting and carbon sequestration. Research priorities include advancing AI agentic systems, LLM as an Abstraction (LLMaaA), AI model optimization and unified abstractions across heterogeneous infrastructure, end-to-end edge-cloud transformation, efficient programming model, middleware and platform, secure infrastructure, application-adaptive cloud systems, and new quantum-classical collaborative workflows. These ideas and solutions encompass both theoretical and practical research questions, requiring coordinated input and support from the research community. #IBM, #ILLINOIS, #AI, #CloudComputing, #HybridCloud, #Sustainability, #QuantumInnovation, #MaterialsDiscovery, #ClimateChange https://lnkd.in/ghcWDAeN

123

2 Comments

I talked about the Journal of Open Source Software (JOSS) today at #force2024, including some comments on how JOSS works with the Principles of Open Scholarly Infrastructure (POSI) as well as issues we've had with scholarly gatekeeping organizations (e.g. Scopus, Web of Science, PubMed Central) slides: https://lnkd.in/guWm5zDj

27

2 Comments

Our new paper, 'Data-Driven Bottleneck Detection on Tehran Highways,' presents a scalable big data solution with global applicability for addressing urban mobility challenges like traffic congestion. Elsevier Iman Gholampour https://lnkd.in/dKgSMqQX

70

2 Comments

More about our new PNAS paper in this interesting article and how merging AI and human efforts can tackle complex mathematical problems 💪 https://lnkd.in/gSVfxqv4

4

Another interesting review on Rotem Elimelech’s paper and this time in Hebrew- check it out at this link: https://lnkd.in/dJPu2Yyc

7

Who is "hallucinating"? Yet another ranking, https://lnkd.in/gAUe3G3k suggests that I am an expert in "Geotechnical engineering, Slope stability analysis, Fluorescence", also something to do with heavy metals. Very honoured to be ranked 157/353 in computer science in Australia, my commiserations to all my colleagues who ranked below me, please learn from my profile, forget about Computer Science, and work on "stability analysis" to have a stable ranking at the top. Hope all good universities follow UZH and leave the hoardings "Best University on the Planet" to the rest: https://lnkd.in/gCF7Mjki

43

7 Comments

New in the AFP: The Category of Sets An axiomatization of category SET on top of HOL, following the presentation given by Halvorson, including Cantor's diagonalization theorem and mathematical induction. https://lnkd.in/g6WTQ-MX

13

I am excited to open up my course on Large Language Models (LLMs) at the University of Pennsylvania to a broader audience through a virtual lecture series, sponsored by Turing.   Join us on Monday, September 23rd, from 1:45 PM - 3:15 PM ET for the first in this series, featuring Yann Dubois, Ph.D. candidate at Stanford University advised by Percy Liang and Tatsu Hashimoto. Yann will talk about "Scalable Evaluation of Large Language Models". Register here: https://lu.ma/or7gm94o Abstract: Evaluation is a cornerstone of machine learning, critical for model development and selection. For LLMs like ChatGPT, evaluation presents unique challenges due to the open-ended nature of their outputs. While human evaluation remains the gold standard, its cost and time-intensive nature make it impractical for rapid development cycles. This talk will provide an overview of scalable approaches to LLM evaluation, focusing on using one LLM to evaluate another. I will discuss the potential benefits and limitations of this approach and explore mitigation strategies for its challenges. This is just the beginning! Stay tuned for more lectures coming online from this course. #ai #machinelearning #largelanguagemodels #llm #upenn #turing

228

6 Comments

Nathan Kwan-Ho Ng presented our paper titled "TailClipper: Reducing Tail Response Time of Distributed Services Through System-Wide Scheduling" at the ACM Symposium on Cloud Computing (ACM SoCC) this week. This work focuses on end-to-end tail latency reduction across distributed services. Read it here: https://lnkd.in/eEQxvGa5 Interesting side note: the paper set an internal record for the highest "tail latency" of a paper from start to finish - took seven years to get it done and published, with many kernel implementation attempts and a few rejections along the way! David Irwin Abel S. Ahmed Ali El Din Don Towsley #cloudcomputing

63

2 Comments