Graduate

Nanotechnology allows for the unique design and functionalization of materials and devices at the nanometer scale for a variety of applications. Our laboratory has synthesized organic and inorganic nanoparticles and nanocomposites for advanced drug delivery, antimicrobial, antifouling, stem cell culture, tissue engineering, and biosensing applications.

We have also fabricated nanofluidic systems for drug screening, in vitro toxicology, clinical sample preparation, and diagnostic applications. The nanosystems allow for the rapid and automated processing of drug candidates and clinical samples in tiny volumes, greatly facilitating drug testing, genotyping assays, infectious disease detection, point-of-care monitoring, as well as cancer diagnosis and prognosis.

In this talk we present two novel ideas: the first is novel method to compute the feedback capacity of channels with memory using reinforcement learning (RL). The second is a new technique of using Graph-based auxiliary random variable to convert a multi-letter expression of feedback capacity formula into a single letter expression.

In RL, one seeks to maximize cumulative rewards collected in a sequential decision-making environment. This is done by collecting samples of the underlying environment and using them to learn the optimal decision rule. The main advantage of this approach is its computational efficiency, even in high dimensional problems. Hence, RL can be used to estimate numerically the feedback capacity of unifilar finite state channels (FSCs) with large alphabet size. The outcome of the RL algorithm sheds light on the properties of the optimal decision rule, which in our case, is the optimal input distribution of the channel.

The insights gained from the RL computation can be converted into analytic, single-letter capacity expressions by solving corresponding lower and upper bounds. The bounds are based on another novel idea of using Graph-based auxiliary random variable

We demonstrate the efficiency of this method by analytically solving the feedback capacity of the well-known Ising channel with a large alphabet. We also provide a simple coding scheme that achieves the feedback capacity.

6:30PM - Check-in
7:00PM - Film Screening Begins
8:40PM - Discussion Begins
9:15PM - Program Ends

From Computer History Museum Events: In a blockbuster paper in 1948, Claude Shannon introduced the notion of a "bit" and laid the foundation for the Information Age. His ideas ripple through such diverse fields as communication, linguistics, genetics, computing, cryptography, neuroscience, artificial intelligence, and cosmology. In later years, he constructed a mathematical theory of juggling, rode unicycles, wrote the first paper on computer chess and built a flaming trumpet!

Join us for a screening of The Bit Player, a film that combines interviews with leading scientists, archival film, inventive animation and compelling commentary from Shannon himself to tell the story of an overlooked genius who revolutionized the world but never lost his childlike curiosity. After the film, there will be a Q & A with the director Mark Levinson (Particle Fever).

Watch the film trailer here.

Microwave Imaging Systems (MIS) have witnessed huge progress during the last few years; thanks to the advances in the semiconductor miniaturization, for both analogue as well as digital circuits. Notwithstanding the ever-increasing demands for imaging solutions serving in the industrial, security, and medical domains. Many efforts are accomplished to move microwave imaging methods from their conventional lab environment to the real applications space. Many more methods are still yet struggling to see the light away from the lab bench and the computer simulators. Collaboration between academia and industry is essential to advance microwave imaging in the benefit of the general public. Monitoring this over the years, it is obvious that educational and research gaps are hindering the progress of this challenging technology. This talk aims to be a first step to raise awareness on the topic and to envision a dedicated research focused initiative in academia.
Microwave imaging has proved to be of a great benefit to many applications and one can surely say that: we just scratched the surface!

Reducing the numerical precision of neural networks is one of the simplest, most effective and most common methods to improve resource efficiency (e.g. by reducing the memory and power requirements). Much research has been invested in finding how to quantize neural nets without significantly degrading performance. I will describe the main bottlenecks and solutions in various settings:
1) 1bit inference (1bit weights and activations), NIPS 2016 - link
2) 8bit training (8bit weights, activations, gradients, and batch-norm), NeurIPS 2018 - link
3) 4bit inference when quantization is done only post-training, Arxiv 2019 - link
4) Calculating the maximum trainable depth as a function of the numerical precision, Arxiv 2019 - link

Given a large cohort of "similar" cases Z1, . . . , ZN one can construct an efficient statistical inference procedure by 'learning from the experience of others' (also known as "borrowing strength" from the ensemble). But what if, instead, we observe a massive database where each case is accompanied with extra contextual information (usually in the form of covariates), making the Zi's non-exchangeable? It's not obvious how we go about gathering strength when each piece of information is fuzzy and heterogeneous. Moreover, if we include irrelevant cases, borrowing information can heavily damage the quality of the inference! This raises some fundamental questions: when (not) to borrow? whom (not) to borrow? how (not) to borrow? These questions are at the heart of the "Problem of Relevance" in statistical inference – a puzzle that remained too little addressed since its inception nearly half a century ago (Efron and Morris, 1971,1972; Efron 2019). The purpose of this talk is to present an attempt to develop basic principles and practical guidelines in order to tackle some of the unsettled issues that surround the relevance problem. Through examples, we will demonstrate how our new statistical perspective answers previously unanswerable questions in a realistic and feasible way.

This is a joint work with my previous student Kaijun Wang, who is currently a postdoctoral research fellow at Fred Hutchinson Cancer Research Center.

As physical robot networks become more pervasive all around us, in the form of teams of autonomous vehicles, fleets of delivery drones, and smart and mobile IoT, it becomes increasingly critical to question the robustness of their coordination algorithms to security threats and/or corrupted data. Indeed, it has been shown that many multi-robot tasks easily fail in the presence of erroneous or hacked data. We investigate the vulnerabilities of important multi-robot algorithms such as consensus, coverage, and distributed mapping to malicious or erroneous data and we demonstrate the potential of communication to thwart certain attacks, such as the Sybil Attack, on these algorithms. Our key insight is that coordinated mobility can be combined with signal processing of communication signals to allow agents to learn important information about the environment and the nature of other agents in the network (for example the presence of cooperative versus adversarial agents). Along these lines, we will present a theoretical and experimental framework for provably securing multi-robot distributed algorithms through careful use of communication. We will present both theoretical results and experimental results on actual hardware implementations for bounding the influence of a Sybil Attack on consensus and on coverage by using observations over the wireless channels. In some cases, we show that the effect of a Sybil Attack can be nearly eliminated with high probability by deriving the appropriate switching function using a sufficient number of observations over the wireless network. Finally, we will briefly describe promising results on new methods for outlier rejection and active rendezvous in a pose graph optimization framework that exploits feedback gathered from communication channels to arrive at improved accuracy.

TBA

Many statistical estimands can expressed as continuous linear functionals of a conditional expectation function. This includes the average treatment effect under unconfoundedness and generalizations for continuous-valued and personalized treatments. In this talk, we discuss a general approach to estimating such quantities: we begin with a simple plug-in estimator based on an estimate of the conditional expectation function, and then correct the plug-in estimator by subtracting a minimax linear estimate of its error. We show that our method is semiparametrically efficient under weak conditions and observe promising performance on both real and simulated data.

This paper establishes the first analytical formula for optimal nonlinear shrinkage of largedimensional covariance matrices. We achieve this by identifying and mathematically exploiting a deep connection between nonlinear shrinkage and nonparametric estimation of the Hilbert transform of the sample spectral density. Previous nonlinear shrinkage methods were numerical: QuEST requires numerical inversion of a complex equation from random matrix theory whereas NERCOME is based on a sample-splitting scheme. The new analytical approach is more elegant and also has more potential to accommodate future variations or extensions. Immediate benefits are that it is typically 1,000 times faster with the same accuracy and accommodates covariance matrices of dimension up to 10,000. The difficult case where the matrix dimension exceeds the sample size is also covered.