# Events

Oct 14

#### Perceptual + Cognitive Biases in Data Visualizations

11:00 AM to 12:30 PM

CSB 480

Cindy Xiong, Northwestern University

Your visual system can crunch vast arrays of numbers at a glance, providing the rest of your brain with critical values, statistics, and patterns needed to make decisions about your data. But that process can be derailed by biases at both the perceptual and cognitive levels. I demonstrate 3 instances of these biases that obstruct effective data communication. First, in the most frequently used graphs – lines and bars – reproductions of average values are systematically underestimated for lines, and overestimated for bars. Second, when people see correlational data, they often mistakenly also see a causal relationship. I’ll show how this error can be even worse for some types of graphs. Third, we’ve all experienced being overwhelmed by a confusing visualization. This may happen because the designer – an expert in the topic – thinks that you’d see what they see. I’ll describe a replication of this real-world phenomenon in the lab, showing that, when communicating patterns in data to others, it is tough for people to see a visualization from a naive perspective. I discuss why these biases happen in our brains, and prescribe ways to design visualizations to mitigate these biases.

Oct 17

#### Rethinking Control

1:00 PM to 2:00 PM

CS conference room (CSB 453)

Linear dynamical systems are a continuous subclass of reinforcement learning models that are widely used in robotics, finance, engineering, and meteorology. Classical control, since the works of Kalman, has focused on dynamics with Gaussian i.i.d. noise, quadratic loss functions and, in terms of provably efficient algorithms, known systems and observed state. In this talk we'll discuss how to apply new machine learning methods to control which relax all of the above: efficient control with adversarial noise, general loss functions, unknown systems, and partial observation.

Based on joint work with Naman Agarwal, Nataly Brukhim, Brian Bullins, Karan Singh, Sham Kakade, Max Simchovitz, and Cyril Zhang.

Oct 18

#### 2,3,…,k: From approximating the number of edges to approximating the number of k-cliques (with a sublinear number of queries)

12:00 PM to 1:00 PM

CS conference room (CSB 453)

Dana Ron, Tel Aviv University

This talk will present an algorithms for approximating the number of k-cliques in a graph when given query access to the graph. This problem was previously studied for the cases of k=2 (edges) and k=3 (triangles). We give an algorithm that works for any k >= 3, and is actually conceptually simpler than the k=3 algorithm.

We consider the standard query model for general graphs via (1) degree queries, (2) neighbor queries and (3) pair queries.

Let n denote the number of vertices in the graph, m the number of edges, and C_k the number of k-cliques.

We design an algorithm that outputs a (1+\epsilon)-approximation (with high probability) for C_k, whose expected query complexity and running time are
O (\frac{n}{C_k^{1/k}}+\frac{m^{k/2}}{C_k}) poly (\log n, 1/\epsilon,k).

Hence, the complexity of the algorithm is sublinear in the size of the graph for C_k = \omega(m^{k/2-1}).

Furthermore, we prove a lower bound showing that the query complexity of our algorithm is essentially optimal (in terms of the dependence on n, m and C_k, up to the dependence on \log n, 1/\epsilon and k).

If time permits, I will also talk shortly about follow-up work on approximate counting of $k$-cliques in bounded-arboricity graphs.

The talk is based on works with Talya Eden and C. Seshadhri.

Oct 21

#### Low-Power Real-Time Edge Computing

11:30 AM to 12:30 PM

CS conference room (CSB 453)

Tulika Mitra, National University of Singapore

Internet of Things (IoT), a network of billion computing devices embedded within physical objects, is revolutionizing our lives. The IoT devices at the edge are primarily responsible only for collecting and communicating the data to the cloud, where the computationally intensive data analytics takes place. However, the data privacy and the connectivity issues - in
conjunction with the fast real-time response requirement of certain IoT applications - call for smart edge devices that should be able to support privacy-preserving, time-sensitive computation for machine intelligence on-site. We will present the computation challenges in edge computing and introduce hardware-software co-designed approaches to overcome these
challenges. We will discuss the design of tiny accelerators that are completely software programmable and can speed up computation to realize the edge computing vision at ultra-low power budget. We will also demonstrate the promise of collaborative computation that engages
heterogeneous processing elements in a synergistic fashion to achieve real-time edge computing.

Oct 25

#### Theory Seminar - Swastik Kopparty

12:00 PM to 1:00 PM

CSB 480 (Computer Science Department)

Swastik Kopparty, Rutgers University

Nov 01

#### Theory Seminar - Piotr Sankowski

12:00 PM to 1:00 PM

CS conference room (CSB 453)

Piotr Sankowski, University of Warsaw

Nov 08

#### Enabling Continuous Learning through Synaptic Plasticity in Hardware

12:00 PM to 1:00 PM

CSB 453

Tushar Krishna, Georgia Tech

Ever since modern computers were invented, the dream of creating artificial intelligence (AI) has captivated humanity. We are fortunate to live in an era when, thanks to deep learning (DL), computer programs have paralleled, and in many cases even surpassed human level accuracy in tasks like visual perception and speech synthesis. However, we are still far away from realizing general-purpose AI. The problem lies in the fact that the development of supervised learning based DL solutions today is mostly open loop. A typical DL model is created by hand-tuning the deep neural network (DNN) topology by a team of experts over multiple iterations, followed by training over petabytes of labeled data. Once trained, the DNN provides high accuracy for the task at hand; if the task changes, however, the DNN model needs to be re-designed and re-trained before it can be
deployed. A general-purpose AI system, in contrast, needs to have the ability to constantly interact with the environment and learn by adding and removing connections within the DNN autonomously, just like our brain does. This is known as synaptic plasticity.

In this talk we will present our research efforts towards enabling general-purpose AI leveraging plasticity in both the algorithm and hardware. First, we will present GeneSys (MICRO 2018), a HW-SW prototype of a closed loop learning system for continuously evolving the structure and weights of a DNN for the task at hand using genetic algorithms, providing 100-10000x higher performance and energy-efficiency over state-of-the-art embedded and desktop CPU and GPU systems. Next, we will present a DNN accelerator substrate called MAERI (ASPLOS 2018), built using light-weight, non-blocking, reconfigurable interconnects, that supports efficient mapping of regular and irregular DNNs with arbitrary dataflows, providing ~100% utilization of all compute units, resulting in 3X speedup and energy-efficiency over our prior work Eyeriss (ISSCC 2016). Finally, time permitting, we will describe our research in enabling rapid design-space exploration and prototyping of hardware accelerators using our dataflow DSL + cost-model called MAESTRO (MICRO 2019).

Nov 11

#### Safe Machine Learning

11:40 AM to 12:40 PM

CS Department 451

Shafi Goldwasser, UC Berkeley

Cryptography and Computational Learning have shared a curious history: a scientific success for one has often provided an example of an impossible task for the other. Today, the goals of the two fields are aligned. Cryptographic models and tools can and should play a role in ensuring the safe use of machine learning. We will discuss this development with its challenges and opportunities.

Host: Jeannette Wing

Nov 13

#### Towards anonymous and metadata private communication at Internet scale

11:40 AM to 12:40 PM

CS Department 451

As the world becomes more connected, privacy is becoming harder to maintain. From social media services to Internet service providers to state-sponsored mass-surveillance programs, many outlets collect sensitive information about the users and the communication between them – often without the users ever knowing about it. In response, many Internet users have turned to end-to-end encryption, like Signal and TLS, to protect the content of the communication. Unfortunately, these works do little to hide the metadata of the communication, such as when and with whom a user is communicating. In scenarios where the metadata are sensitive, encryption alone is not sufficient to ensure users’ privacy.

Most prior systems that provide metadata private communication fall into one of two categories: systems that (1) provide formal privacy guarantees against global adversaries but do not scale to large numbers of users, or (2) scale easily to a large user base but do not provide strong guarantees against global adversaries. I will present two systems that aim to bridge the gap between the two categories to enable private communication with strong guarantees for many millions of users: Quark, a horizontally scalable anonymous broadcast system that defends against a global adversary who monitors the entire network and controls a significant fraction of the servers, and Crossroads, which provides metadata private communication between two honest users against the same adversary using a novel cryptographic primitive.

This talk is based on Albert Kwon's recently completed MIT PhD dissertation.

Nov 25

#### From Ethical Challenges of Intelligent Systems to Embedding Ethics in Computer Science Education

11:40 AM to 12:40 PM

CSB 451

Barbara Grosz, Harvard University

Computing technologies have become pervasive in daily life, sometimes bringing unintended but harmful consequences. For students to learn to think not only about what technology they could create, but also whether they should create that technology and to recognize the ethical considerations that should constrain their design, computer science curricula must expand to include ethical reasoning about the societal value and impact of these technologies. This talk will describe Harvard's Embedded EthiCS initiative, a novel approach to integrating ethics into computer science education that incorporates ethical reasoning throughout courses in the standard computer science curriculum. It changes existing courses rather than requiring wholly new courses. The talk will begin with a short description of my experiences teaching the course "Intelligent Systems: Design and Ethical Challenges" that inspired the design of Embedded EthiCS. It will then describe the goals behind the design, the way the program works, lessons learned and challenges to sustainable implementations of such a program across different types of academic institutions.
Host: Augustin Chaintreau

Dec 02

#### Distinguished Lecture - Bill Freeman

11:40 AM to 12:40 PM

CS Department 451

Bill Freeman, MIT