Electrical Engineering & Computer Sciences
http://events.berkeley.edu/index.php/calendar/sn/eecs.html
Upcoming EventsBLISS Seminar: Decoding from Random Histogram Queries, Oct 17
http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103548&date=2016-10-17
Consider a population consisting of n individuals, each of whom has one of d types (e.g. their blood type, in which case d=4). We are allowed to query the database by specifying a subset of the population, and in response we observe a noiseless histogram (a d-dimensional vector of counts) of types of the pooled individuals. This measurement model arises in practical situations such as pooling of genetic data and may also be motivated by privacy considerations. We are interested in the number of queries one needs to unambiguously determine the type of each individual, first ignoring computational considerations, and then taking them into account. <br />
We study the power of random queries, where in each query, we choose a random subset of individuals of size proportional to n. We discuss the information-theoretic question and sketch the key steps of a rigorous proof of almost matching upper and lower bounds on the minimum number of queries m such that there is no other solution than the planted one. Time permitting, we discuss computational considerations, and give heuristic arguments that an AMP-based algorithm breaks down way above the information-theoretic threshold, indicating the existence of a wide “detectable but hard” region in the phase space of the problem. <br />
<br />
The proof relies on an exact computation of the ``annealed free energy" of the model in the thermodynamic limit. To this end, we investigate the rich combinatorial structure of the problem, and as a by-product of the analysis, we show a curious identity that relates the Gaussian integral over the space of Eulerian flows of a graph to its spanning tree polynomial. <br />
<br />
This is joint work with Aaditya Ramdas, Florent Krzakala, Lenka Zdeborova, and Michael Jordan.http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103548&date=2016-10-17Solid State Technology and Devices Seminar, Oct 21
http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103527&date=2016-10-21
Two-dimensional (2D) atomic crystals are recently discovered materials that are only atoms thick, and yet can span laterally over millimeters. The diverse family of such materials includes graphene: a semimetal with massless relativistic charge carriers, and monolayer transition metal dichalcogenides (TMDCs), such as molybdenum disulfide (MoS2): direct band gap semiconductors with strong spin-orbit interaction. Since every atom in these materials belongs to the surface, their physical properties are greatly affected by the immediate environment. In my talk, I will demonstrate that ultra-high electronic and optical quality of 2D atomic crystals can be obtained by tuning the local microenvironment, and I will discuss device applications. <br />
In the first part of the talk, I will demonstrate that the electronic quality of graphene is enhanced when placed into a high dielectric stationary liquid environment, through suppression of Columbic scattering strength. I will demonstrate the use of graphene field effect transistors (FETs) in sensing different physical parameters of nanometer-thick interfacial liquid volumes, both stationery and moving. By embedding graphene FETs in a microfluidic channel, I will demonstrate sensing of flow velocity – with sensitivity 70nL/min, and ion concentration with sensitivity as low as 40 nM. Overall, our results highlight the usefulness of graphene FETs for applications in ultra-precise fluidic sensing and as a potential replacement for silicon in next generation transistors. <br />
In the second part of my talk, I will focus on mononalyer TMDCs and explore the formation, binding energies, and dissociation mechanisms of various excitons in monolayer TMDCs through photocurrent spectroscopy and photoluminescence measurements. I will also demonstrate that their optical properties, fluorescence quantum yield, and transparency can be tuned via electrical gating. Our findings suggest the possibility of TMDCs for diverse applications ranging from nanoscale electro-optical modulators to novel energy harvesting devices.http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103527&date=2016-10-21Stimuli-Responsive Smart Soft Materials, Oct 21
http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=100705&date=2016-10-21
Machine technology frequently puts magnetic or electrostatic repulsive forces to practical use, as in maglev trains, vehicle suspensions, or non-contact bearings. In contrast, materials design overwhelmingly focuses on attractive interactions, such as in the many advanced polymer-based composites, where inorganic fillers interact with a polymer matrix to improve mechanical properties. <br />
<br />
However, articular cartilage strikingly illustrates how electrostatic repulsion can be harnessed to achieve unparalleled functional efficiency: it permits virtually frictionless mechanical motion within joints, even under high compression. <br />
<br />
Here we describe a composite hydrogel with anisotropic mechanical properties dominated by electrostatic repulsion between negatively charged unilamellar titanate nanosheets embedded within it. Crucial to the behaviour of this hydrogel is the serendipitous discovery of cofacial nanosheet alignment in aqueous colloidal dispersions subjected to a strong magnetic field, which maximizes electrostatic repulsion and thereby induces a quasi-crystalline structural ordering over macroscopic length scales and with uniformly large face-to-face nanosheet separation. <br />
<br />
We fix this transiently induced structural order by transforming the dispersion into a hydrogel using light-triggered in situ vinyl polymerization. The resultant hydrogel, containing charged inorganic structures that align cofacially in a magnetic flux, deforms easily under shear forces applied parallel to the embedded nanosheets yet resists compressive forces applied orthogonally. <br />
<br />
We anticipate that the concept of embedding anisotropic repulsive electrostatics within a composite material, inspired by articular cartilage, will open up new possibilities for developing soft materials with unusual functions.<br />
<br />
More recently, we reported a novel carbon nitride polymer, which shows anomalous mechanical responses to minute fluctuations in the ambient humidity.<br />
<br />
This visiting lecture is generously sponsored by Sigma-Aldrich Corp., a Merck company.http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=100705&date=2016-10-21Dissertation Talk: New Tools for Econometric Analysis of High-Frequency Time Series Data, Oct 21
http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103365&date=2016-10-21
This is a dissertation talk in partial requirement for the completion of a PhD in the EECS department. Advisor: Claire Tomlinhttp://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103365&date=2016-10-21BLISS Seminar: Data-driven methods for sparse network estimation, Oct 24
http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103549&date=2016-10-24
A Graphical model is a probabilistic model for which a graph is used to represent the conditional independence between random variables. Such models have become extremely popular tools for modeling complex real-world systems. Learning graphical models is of fundamental importance in machine learning and statistics and is often challenged by the fact that only a small number of samples are available relative to the number of variables. Several methods (such as Graphical Lasso) have been proposed to address this problem. However, there is a glaring lack of concrete case studies that clearly illustrate the limitations of the existing computational methods for learning graphical models. In this talk, I will propose a circuit model that can be used as a platform for testing the performance of different statistical approaches. I will show that the data generated from this circuit model exhibits similar trends to resting-state functional MRI (fMRI) data, and then discuss how our findings from case studies of the circuit model can be used to study graphical models for brain functional connectivity from fMRI data. Motivated by the above results, I will develop new insights into regularized semidefinite program (SDP) problems by working through the Graphical Lasso algorithm. Graphical Lasso is a popular method for learning the structure of a Gaussian model, which relies on solving a computationally-expensive SDP. I will derive sufficient conditions under which the solution of this large-scale SDP has a simple formula, and test them on electrical circuits and fMRI data.http://events.berkeley.edu/index.php/calendar/sn/eecs.html?event_ID=103549&date=2016-10-24