Tomorrow, at the Seminar in Advances in Computing, we have an exciting talk by Dr. Laura Boccanfuso who is the founder and CEO of Van Robotics. The talk will be focused on AI- and robot-assisted learning for students. Dr. Boccanfuso is a UofSC alumna who received her doctoral degree under the supervision of Dr. Jason O'Kane.

Meeting Location:
Storey Innovation Center 1400
 

BIO: Dr. Laura Boccanfuso is founder and CEO of Van Robotics, a social robotics company headquartered in Columbia, SC that builds AI-enabled tutors. Laura received her PhD in Computer Science and Engineering from the University of South Carolina and later worked in the Yale University Social Robotics Lab and Child Study Center a Postdoctoral researcher and Associate Research Scientist. She completed the Yale Venture Creation Program in 2016, officially launched the company in 2017 and was selected for the Techstars Accelerator in Austin later that year. Laura’s work focuses on wholistic robot-assisted learning, incorporating best practices in human-robot interaction that leverage sound educational pedagogy, cognitive and learning science and machine learning techniques to accelerate learning.

TALK ABSTRACT: Existing AI-enabled applications that personalize learning for students primarily focus on assessing which skill(s) the student has mastered, which skills they have not yet mastered, and the optimal learning trajectory, or ordered series of skills building lessons, that will accelerate skills mastery for the individual. However, the process of learning effectively is not confined to a well-defined set of cognitive leaps that result in skills mastery. Instead, effective learning is very often driven by a set of personal factors that includes individual cognitive ability, mental focus, social and emotional intelligence, and intrinsic or extrinsic motivation. In this presentation, we will explore some methodologies for actively collecting and processing measures of these traits and conditions, and discuss some of the constraints that must be addressed in order to implement an effective multi-modal approach for highly personalized learning.

See this ACM Event Page for details.

This Thursday I will give an introduction to the capacity of Django to make professional websites with ease. I'll explain the basics of starting your own site, how Django works, and finally go into some of the work I am doing with Django. Since this will involve a mix of Python, JavaScript, HTML, and CSS, we will also briefly touch on anything that may be new.

Sep 21, Tuesday, 10:00-11:15 am 

Blackboard link: https://us.bbcollab.com/guest/f567247c101145cebc6eaa937af2cecd 

 Sep 23, Tuesday, 10:00-11:15 am 

Blackboard link: https://us.bbcollab.com/guest/f567247c101145cebc6eaa937af2cecd  

On campus class at Seminar Room, AI Institute, 1112 Greene St, Columbia (5^th Floor; Science & Technology Building) 

Speaker Bio: 

Dr. Diptikalan Saha (Dipti) is a Senior Technical Staff Member and manager of Reliable AI team in Data&AI department of IBM Research at Bangalore. His research interest includes Artificial Intelligence, Natural Language Processing, Knowledge representation, Program Analysis, Security, Software Debugging, Testing, Verification, and Programming Languages. He received a  Ph.D. degree in Computer Science from the State University of New York at Stony Brook his B.E. degree in Computer Science and Engineering from Jadavpur University. His group’s work on Bias in AI Systems is available through AI OpenScale in IBM Cloud as well as through open-source AI Fairness 360.  

Related Material 

• Blogs:     
  • https://medium.com/@diptikalyan?p=5ce7347f5f75 
  • https://www.ibm.com/blogs/watson/2021/06/trustworthy-ai-assessment-mitigation/ 
  • https://www.ibm.com/blogs/watson/2020/10/how-ibm-makes-ai-based-on-trust-fairness-and-explainability/ 
•  Surveys:     
  • Fairness: https://arxiv.org/pdf/1908.09635.pdf     
  • Explainability: https://christophm.github.io/interpretable-ml-book/ 
  • AI Testing: https://www.researchgate.net/publication/334048996_Machine_Learning_Testing_Survey_Landscapes_and_Horizons 
  • Counterfactual: https://arxiv.org/abs/2010.10596 
•  Tools: 
  • AIF360: https://aif360.mybluemix.net/ 
  • AIX360: https://aix360.mybluemix.net/ 

Sep 21, Tuesday, 10:00-11:15 am 

Blackboard link: https://us.bbcollab.com/guest/f567247c101145cebc6eaa937af2cecd 

 Sep 23, Tuesday, 10:00-11:15 am 

Blackboard link: https://us.bbcollab.com/guest/f567247c101145cebc6eaa937af2cecd  

On campus class at Seminar Room, AI Institute, 1112 Greene St, Columbia (5^th Floor; Science & Technology Building) 

Speaker Bio: 

Dr. Diptikalan Saha (Dipti) is a Senior Technical Staff Member and manager of Reliable AI team in Data&AI department of IBM Research at Bangalore. His research interest includes Artificial Intelligence, Natural Language Processing, Knowledge representation, Program Analysis, Security, Software Debugging, Testing, Verification, and Programming Languages. He received a  Ph.D. degree in Computer Science from the State University of New York at Stony Brook his B.E. degree in Computer Science and Engineering from Jadavpur University. His group’s work on Bias in AI Systems is available through AI OpenScale in IBM Cloud as well as through open-source AI Fairness 360.  

Related Material 

• Blogs:     
  • https://medium.com/@diptikalyan?p=5ce7347f5f75 
  • https://www.ibm.com/blogs/watson/2021/06/trustworthy-ai-assessment-mitigation/ 
  • https://www.ibm.com/blogs/watson/2020/10/how-ibm-makes-ai-based-on-trust-fairness-and-explainability/ 
•  Surveys:     
  • Fairness: https://arxiv.org/pdf/1908.09635.pdf     
  • Explainability: https://christophm.github.io/interpretable-ml-book/ 
  • AI Testing: https://www.researchgate.net/publication/334048996_Machine_Learning_Testing_Survey_Landscapes_and_Horizons 
  • Counterfactual: https://arxiv.org/abs/2010.10596 
•  Tools: 
  • AIF360: https://aif360.mybluemix.net/ 
  • AIX360: https://aix360.mybluemix.net/ 

This week at the Seminar in Advances in Computing, we have an exciting talk by Dr. Mohammed Zahran from New York University. The talk will be focused on machine learning and adaptive computer architectures. Please use the below link if you are interested in joining the meeting virtually. Also, please share it with your students if you believe attending the talk will benefit them.

Meeting Location:
Storey Innovation Center 1400

Meeting Link (click me)

BIO: Mohamed Zahran is a professor with the Computer Science Department at Courant Institute of NYU. He received his Ph.D. in Electrical and Computer Engineering from University of Maryland at College Park. His research interests span several aspects of computer architecture, such as architecture of heterogeneous systems, hardware/software interaction, and biologically-inspired architectures.  He served in many review panels in organizations such as DoE and NSF, served as PC member in many premiere conferences and as reviewer for many journals.  He will serve as the general co-chair of the 49th International Symposium of Computer Architecture (ISCA) to be held in NYC in 2022. Zahran is an ACM Distinguished Speaker, a  senior member of IEEE, a senior member of ACM,  and a member of Sigma Xi scientific honor society.

TALK ABSTRACT: For the past decade people have been predicting the death of Moore's law, which is now as likely as ever both for technological reasons and for economical reasons. We have already kissed Dennard's scaling goodbye since around 2004. Given that, we are left with two options to get performance: parallel processing and specialized domain specific architectures (e.g. GPUs, TPUs, FPGAs, ...). But programs have different characteristics that change during the lifetime of a program's execution and across different programs. Both general purpose multicore and specialized architectures are designed with an average case in mind, whether this average case is for a general purpose parallel program or a specialized program like GPU-friendly ones,  and this is anything but efficient in an era where power efficiency is as important as performance. Building a machine that can adapt to program requirements is necessary but not sufficient. In this talk we discuss the research path toward building a machine that learns from executing different programs and adapts for best performance when executing unseen programs. We discuss the main challenges involved: a fully adaptable machine is expensive in terms of hardware requirement and transferring learning from a program execution to another program execution is challenging.

Live Virtual Meeting Link

Meeting Location:
Storey Innovation Center 1400
 

BIO: Dr. Qi Zhang is an assistant professor of the Computer Science and Engineering department and the Artificial Intelligence Institute at the University of South Carolina. He got his Ph.D. from the Computer Science and Engineering department at the University of Michigan. His research aims for solutions for coordinating systems of decision-making agents operating in uncertain, dynamic environments. As hand-engineered solutions for such environments often fall short, He uses ideas from planning and reinforcement learning to develop and analyze algorithms that autonomously coordinate agents in an effective, trustworthy, and communication-efficient manner. In particular, He has been working on social commitments for trustworthy coordination, communication learning, and language emergence among coordinated agents and applications of (multi-agent) reinforcement learning such as intelligent transportation systems, dialogue systems, etc.

ABSTRACT: This talk will present two recent works on training homogeneous reinforcement learning (RL) agents in two distinct scenarios, respectively. The first scenario considers training a group of homogeneous agents that will be deployed in isolation to perform a single-agent RL task, which finds applications in ensemble RL. The first work develops effective techniques for training such as an ensemble of deep Q-learning agents, which help achieve state-of-the-art policy distillation performance in Atari games and continuous control tasks. The second scenario considers training a group of homogeneous agents to cooperatively perform a multi-agent RL task such as team sports. The second work develops novel techniques that exploit the homogeneity to train the agents in a distributed and communication-efficient manner.

See https://www.icaslab.com/Seminar

Meeting Link:
https://teams.microsoft.com/l/meetup-join/19%3ameeting_YWQ4NzcyYjMtYmZjNC00MTNjLTk4NTItYjRkOTFmYjk3NTRk%40thread.v2/0?context=%7b%22Tid%22%3a%224b2a4b19-d135-420e-8bb2-b1cd238998cc%22%2c%22Oid%22%3a%225fc2170a-7068-4a33-9021-df11b94ba696%22%7d

BIO: Dr. Xiaochen Guo is an associate professor in the Department of Electrical and Computer Engineering at Lehigh University. Dr. Guo received her Ph.D. degree in Electrical and Computer Engineering from the University of Rochester, and received the IBM Ph.D. Fellowship twice. Dr. Guo’s research interests are in the broad area of computer architecture, with an emphasis on leveraging emerging technologies to build energy-efficient microprocessors and memory systems. Dr. Guo is an IEEE senior member and a recipient of the National Science Foundation CAREER Award, the P. C. Rossin Assistant Professorship, and the Lawrence Berkeley National Laboratory Computing Sciences Research Pathways Fellowship.

ABSTRACT: General-purpose computing systems employ memory hierarchies to provide the appearance of a single large, fast, and coherent memory for general applications with good locality. However, conventional memory hierarchies cannot provide sufficient isolation for security workloads, support richer semantics, or hide memory latency for irregular memory accesses. This talk will present two of our recent works aiming to address these special needs in important workloads. In the first work, we propose to add a virtually addressed, set-associative scratchpad (SPX64) to a general-purpose CPU to support isolation and hash lookups in security and persistent applications. The proposed scratchpad is placed alongside of a traditional cache, and is able to avoid many of the programming challenges associated with traditional scratchpads without sacrificing generality. SPX64 delivers increased security and improves performance. In the second work, a software-assisted hardware prefetcher is proposed, which focuses on repeating irregular memory access patterns for data structures that cannot benefit from conventional memory hierarchies and hardware prefetchers. The key idea is to provide a programming interface to record cache miss sequence on the first appearance of a memory access pattern and prefetch through replaying the pattern on the following repetitions. By leveraging the programmer knowledge, the proposed Record-and-Replay (RnR) prefetcher can achieve over 95% prefetching accuracy and miss coverage.

DISSERTATION DEFENSE

Department of Computer Science and Engineering

University of South Carolina

 Towards More Trustworthy Deep Learning: Accurate, Resilient, and Explainable Countermeasures Against Adversarial Examples

Author : Fei Zuo

Advisor : Dr. Qiang Zeng

Date : Aug 12, 2021

Time : 11:00am

Place : Virtual Defense

Abstract

Despite the great achievements made by neural networks on tasks such as image classification, they are brittle and vulnerable to adversarial example (AE) attacks. Along with the prevalence of deep learning techniques, the threat of AEs attracts increasingly attentions since it may lead to serious consequences in some vital applications such as disease diagnosis. 

To defeat attacks based on AEs, both detection and defensive techniques attract the research community’s attention. While many countermeasures against AEs have been proposed, recent studies show that the existing detection methods usually goes ineffective when facing adaptive AEs. In this work, we exploit AEs by identifying their noticeable characteristics.

First, we noticed that L2 adversarial perturbations are among the most effective but difficult-to-detect attacks. How to detect adaptive L2 AEs is still an open question. At the same time, we find that, by randomly erasing some pixels in an L2 AE and then restoring it with an inpainting technique, the AE, before and after the steps, tends to have different classification results, while a benign sample does not show this symptom. We thus propose a novel AE detection technique, Erase-and-Restore (E&R), that exploits the intriguing sensitivity of L2 attacks. Comprehensive experiments conducted on standard image datasets show that the proposed detector is effective and accurate. More importantly, our approach demonstrates strong resilience to adaptive attacks. We also interpret the detection technique through both visualization and quantification.

Second, previous work considers that it is challenging to properly alleviate the effect of the heavy corruptions caused by L0 attacks. However, we argue that the uncontrollable heavy perturbation is

an inherent limitation of L0 AEs, and thwart such attacks. We thus propose a novel AE detector by converting the detection problem into a comparison problem. In addition, we show that the pre-processing technique used for detection can also work as an effective defense, which has a high probability of removing the adversarial influence of L0 perturbations. Thus, our system demonstrates not only high AE detection accuracies, but also a notable capability to correct the classification results.

Finally, we propose a comprehensive AE detector which systematically combines the two detection methods to thwart all categories of widely discussed AEs, i.e., L0, L2, and L∞ attacks. By acquiring the both strengths from its assembly components, the new hybrid AE detector is not only able to distinguish various kinds of  

AEs, but also has a very low false positive rate on benign images. More significantly, through exploiting the noticeable characteristics of AEs, the proposed detector is highly resilient to adaptive attack, filling a critical gap in AE detection.

When: Tuesday, April 27, 11:00-12:00

https://zoom.us/j/98717098004?pwd=T3lPSGZ2K1pKTllQaWhRMDBtNVcrUT09

Speaker: Dr. Christian O’Reilly, McGill University, Canada

Talk abstract: Modeling is the bedrock on which science and technology have been built. Nowadays, almost every part of manufactured objects – may it be a supercomputer or a simple light bulb -- is modeled and simulated for us to gain a comprehensive understanding of how it works and how it will react under different conditions. Compared to human-made objects, our ability to get a grip on complex biological systems such as the brain has been hindered by these systems being black boxes which inner workings were mostly unknown. As we gain more insights on the mechanisms at play, our capacity to model and simulate these systems increases and further shed light on their remaining mysteries. In parallel, as the advances in medicine and science provide us with a finer appreciation of these biological systems, it also generates more intricate challenges. Tackling these new problems often requires integrating many sources of knowledge across fields and scales, from slow-evolving social factors to millisecond molecular interactions. Understanding complex multi-factorial and multidimensional neurodevelopmental issues like those present in the autistic spectrum disorder is such a problem. In this context, setting up a solid analytical framework empowered by modeling and simulation is even more important. In the first half of this talk I will go over some of my experiences in analyzing and modeling neuronal systems at different scales, from the macroscopic whole-brain scale to the microscopic cellular scale. Then, in the second part, building on these experiences I will make a case for the importance of systematically benchmarking the different aspects of the brain across scales and integrating such knowledge into analytical tools that we can use for scientific discoveries and clinical decisions. Speaker bio (short): Christian O’Reilly (Google Scholar) received his B.Ing (electrical eng.; 2007), his M.Sc.A. (biomedical eng.; 2011), and his Ph.D. (biomedical eng.; 2012) from the École Polytechnique de Montréal where he worked under the mentoring of Pr. R. Plamondon to apply pattern recognition and machine learning to predict brain stroke risks. He was later a postdoctoral fellow in Pr. T. Nielsen’s laboratory at the Center for Advanced Research in Sleep Medicine of the Hôpital du Sacré-Coeur/Université de Montréal (2012-2014) and then a NSERC postdoctoral fellow at McGill's Brain Imaging Center (2014-2015) where he worked in Pr. Baillet’s laboratory on characterizing EEG sleep transients, their sources, and their functional connectivity. During this period, he also was a visiting scholar in Pr. K. Friston's laboratory at the University College of London to study effective connectivity during sleep transients using dynamic causal modeling, an approach based on the Bayesian inversion of neural mass models. He later took on a 6-month fellowship with the Pr. M. Elsabbagh on functional connectivity in autism after which he moved to Switzerland to work for the Blue Brain project (Pr. S. Hill; EPFL; 2015-2018) where he led efforts on large-scale biophysically detailed modeling of the thalamocortical loop. Since 2020, he resumed his collaboration with the Dr. Elsabbagh as a research associate at the Azrieli Centre for Autism Research (McGill) where he is studying brain connectivity in autism and related neurodevelopmental disorders.