Student Speakers
Session 1
Abstract:
Carbon nanotube network (CNN) devices are useful in integrated circuits and display drivers, particularly in applications that make use of thin film transistors (TFTs) on flexible or transparent substrates. However, the performance of CNN devices is usually limited by the high electrical and thermal resistances at the individual nanotube junctions (NJs). In this study, we present a novel method to improve such resistances by depositing metal at highly resistive NJs using a localized chemical vapor deposition (CVD) process. By passing current through the devices, we induce nanometer scale heating at the NJs. This is done in the presence of metal CVD precursors in a vacuum environment, enabling the selective deposition of metal to nanosolder the NJs. The effectiveness of this nanosoldering process depends on the metal work function, and it can improve the on/off current ratio of a CNN device by nearly an order of magnitude. This technique could also be applied to other device types where nanoscale resistance components limit overall device performance.
Abstract:
We study the sum-capacity of information stable memoryless real-valued additive white Gaussian noise interference channels (IC), a class which includes the classical IC, the block asynchronous IC, and the IC with partial codebook knowledge as special cases. We develop and analytically characterize the rates achievable by a new strategy that uses superpositions of Gaussian and discrete random variables as channel inputs and does not require joint decoding of intended and interfering messages. By using such inputs, the block asynchronous IC and the IC with partial codebook knowledge are shown to have: (i) no performance loss compared to the classical IC (synchronous and with full codebook knowledge) in terms of generalized degrees of freedom, and (ii) can achieve to within an additive gap of O(1) or O(loglog(SNR)) to the symmetric sum-capacity of the classical IC.
Abstract:
This paper considers the K-user cognitive interference channel with one primary and K-1 secondary/cognitive transmitters with a cumulative message sharing structure, i.e., cognitive transmitter i, i in [2:K], non-causally knows all messages of the users with index less than i. We first propose a computable outer bound valid for any memoryless channel and show the sum-rate to be achievable for the symmetric K-user Linear Deterministic Channel. Interestingly, for the K-user channel having only the K-th transmitter know all other messages is sufficient to achieve the sum-capacity, i.e., cognition at transmitters 2 to K-1 is not needed. Next, the sum-capacity of the symmetric Gaussian noise channel is characterized to within a constant additive and multiplicative gap, which depend on K. The achievable scheme for the additive gap is based on Dirty Paper Coding and can be thought of as a MIMO-broadcast scheme where only one encoding order is possible due to the message sharing structure. As opposed to other interference channel models, a single scheme suffices for both the weak and strong interference regimes. Moreover it is only required for transmitters 2 to K-1 to have, in addition to their own message, non-causal message knowledge of the transmitter 1’s message.
____________________________________________________________________
Abstract:
Most wireless communication networks are two-way, where nodes act as both sources and destinations of messages. This allows for “adaptation” at or “interaction” between the nodes – a node’s channel inputs may be functions of its message(s) and previously received signals allowing for potentially larger rates than those achievable in feedback-free one-way channels where inputs are functions of messages only. However, examples exist of channels where adaptation is not beneficial from a capacity perspective; we ask whether analogous results hold for several multi-user two-way networks. We first consider deterministic two- way channel models: the binary modulo-2 addition channel and a generalization of this, and the linear deterministic channel which models Gaussian channels at high SNR. For these deterministic models we obtain the capacity region for the two-way multiple access/ broadcast channel, the two-way Z channel and the two-way interference channel (under certain “partial” adaptation constraints in some regimes). We permit all nodes to adapt their channel inputs to past outputs (except for portions of the linear high-SNR two-way interference channel where we only permit 2 of the 4 nodes to fully adapt). However, we show that the two-way fully or partially adaptive capacity region consists of two parallel “one-way” regions operating simultaneously in opposite directions, i.e. adaptation is useless. We next consider two noisy channel models: first, the Gaussian two-way MAC/BC, where we show that adaptation can at most increase the sum-rate by 1/2 bit in each direction. Next, for the two-way interference channel, partial adaptation is shown to be useless when the interference is very strong. In the strong and weak interference regimes, we show that the non-adaptive Han and Kobayashi scheme utilized in parallel in both directions achieves to within a constant gap for the symmetric rate of the fully (for some regimes) or partially (for the remaining regimes) adaptive models. The central technical contribution is the derivation of new, computable outer bounds which allow for adaptation.
Session 2
Abstract:
A major obstacle in computed tomography (CT) is the reduction of harmful x-ray dose while maintaining the quality of reconstructed images. Methods which exploit the sparse representations of tomographic images have long been known to improve
the quality of reconstructions from low-dose data. Recent work has shown the promise of adaptive, rather than fixed, sparse representations. In particular, the synthesis dictionary learning framework has been shown to outperform traditional regularization techniques. However, these methods scale poorly with data size, and may be prohibitively expensive for practical tomographic reconstruction. We propose a new method for image reconstruction from low-dose data by combining the statistical iterative reconstruction framework with an adaptive sparsifying transform penalty. An alternating minimization approach is used to jointly reconstruct the image while learning a sparsifying transform adapted to the particular image being reconstructed. The Alternating Direction Method of Multipliers is used to provide a computationally efficient solution to the statistically weighted minimization problem. Numerical experiments demonstrate that adaptive sparsifying transform regularization outperforms state-of-the-art synthesis sparsity methods at speeds rivaling total-variation regularization.
Abstract: Lower-limb prosthetic devices are often controlled by impedance control. This approach requires choosing many impedance controller parameters. In the past, clinicians have chosen these parameters by trial and error, an expensive process that takes about four hours for each patient. I will describe a new way to automate the tuning of these parameters. Our approach depends on a method of learning from demonstration called inverse optimal control. This learning algorithm is used to characterize “invariant outputs” of human locomotion given observations of unimpaired human walkers, and subsequently learn prosthetic controller parameters automatically.
Abstract:
Abstract:
Abstract:
Abstract:
Abstract:
In the classical quickest change detection problem, the objective is to detect a change in the
distribution of a sequence of random variables with minimum possible delay subject to a constraint
on the false alarm rate. We consider this problem with an additional constraint on the cost of
observations used before the change occurs. This problem is encountered in many engineering
application. We propose a minimax formulation for this problem. For the case when the pre- and
post-change distributions are known, we show that a two-threshold generalization of the classical
single threshold test is asymptotically optimal. We extend this theory to sensor networks and to the
case when the post-change distribution is unknown.
Session 3
Abstract:
Wearable technology is a category of consumer electronics including devices such as Google Glass and Samsung Galaxy Gear. Majority of these devices are physical activity trackers, e.g., Nike+ Fuelband, Jawbone UP, and Fitbit Force. These devices include motion sensors such as accelerometers and are worn by the user on his or her wrist. These devices distinguish and track the activities performed by the user during the day. For example, the number of steps, the number of stairs climbed, or the number of hours spent doing intense physical activity.
In this talk, I will demonstrate a physical activity recognition system to learn, recognize, and distinguish general types of physical activities using sensor data from a wearable device. I will briefly discuss the nonlinear filtering algorithms behind this as well as many use-cases for this technology. I will demonstrate the technology with a live demo, and show two video clips: A runner using this technology to track walking, jogging, and running; and a weight-lifter using it to track different weight-lifting exercises.
Abstract:
Over the past few years the need has grown for low-cost, high-speed, and accurate biomolecule sensing technology. Graphene is a promising choice for use in such sensing applications, as its single-atom thickness and unique electronic structure is suitable for probing biomolecules like DNA at a very high resolution. We propose the design of a transistor containing a graphene nanoribbon sensing layer with a nanopore for the simultaneous detection and control of a translocating DNA molecule. Through the combination of molecular dynamics simulations, a self-consistent Poisson equation solver, and electronic transport theory, we show that the motion of a DNA molecule through a nanopore can be observed by measuring conductance modulations in the graphene nanoribbon. We also demonstrate that the sensitivity of the graphene sheet conductance to external charges can be enhanced by modulating its carrier concentration as well as by choosing a quantum point contact geometry for the graphene nanoribbon. In addition, we propose the use of extra gates to control both the lateral and translocating motion of a DNA
molecule inside the nanopore.
Abstract:
Abstract:
Session 4
Abstract:
Abstract:
Abstract:
Abstract:
Posters
Abstract:
Although chemical modification of graphene creates a band gap, achieving thermal
and chemical stability in fluorinated or hydrogenated graphene remains challenging. Band
gap engineering through size confinement with graphene nanoribbons suffers from serious
impediments to device fabrication. Graphene antidot lattices are an elegant alternative to
graphene nanoribbons that have the potential to be easily transferred onto device surfaces.
Experimental work on graphene antidot lattices has been limited to top-down fabrication methods
which do not reach size scales necessary for significant band gaps. Bottom-up approaches
have produced highly ordered polyphenylene networks, but not porous graphene. We examine
the formation of graphene antidot lattices through the on-surface polymerization of halogenated
aromatic molecules. In addition to thermally mediated self-assembly, tip-induced assembly offers
a potential route for directing nanostructure formation. We deposit 1,3,5-tris(2-bromophenyl)
benzene (TBB) onto gold. We characterize the surfaces using scanning tunneling microscopy
(STM) and scanning tunneling spectroscopy (STS). Samples prepared with the substrate held at
room temperature during deposition appear streaky when imaged with STM, indicating that at room
temperature the molecule is highly mobile on gold, and has not polymerized. With the substrate
held above 200° C during deposition, TBB nucleates into disordered networks at step edges. When
areas imaged as streaky are exposed to high energy tunneling electrons, a disordered network
appears underneath the path of the electron beam. Depositing the molecule onto a substrate
heated at 400°C and above leads to extended networks that span entire terraces. High index
facets show highly ordered regions which we attribute to close-packed bromine adatom islands.
A disordered porous network assembles on low index facets. By examining the role of substrate
temperature in the deposition of 1,3,5-tris(2-bromophenyl benzene) on gold, we have found
conditions that lead to a porous structure. We also show evidence of tip-induced polymerization of
halogenated aromatic molecules.
SMA-STM utilizes a backside illumination technique to reduce tip heating effects. The evanescent wave of a laser undergoing total internal reflection excites molecules on the surface, thus changing their local density of states. The excitation laser is amplitude modulated, allowing for simultaneous detection of the STM current and its derivative by a lock-in amplifier. 15 nm thick platinum gold hybrid films deposited by electron beam evaporation onto c-plane sapphire substrates serve as substrates. SMA-STM performed on quantum dots and carbon nanotubes deposited by dry contact transfer onto a PtAu film, resulted in a strong, phase dependent absorption signal.
In standard clustering problems, data points are represented by vectors, and by stacking them together, one forms a data matrix with row or column cluster structure. We consider a class of binary matrices which exhibit both row and column cluster
structure, and our goal is to exactly recover the underlying row and column clusters by observing only a small fraction of noisy entries. We first derive a lower bound on the minimum number of observations needed for exact cluster recovery. Then we propose three algorithms with different runtime and compare the number of observations needed by them for successful cluster recovery. Our analytical results show smooth time-data trade-offs: one can gradually reduce the computational complexity when increasingly more observations are available.
____________________________________________________________________
____________________________________________________________________
Abstract: Learning is a process that occurs over a lifetime. Yet, even after many lifetimes, the mechanics of how it is realized in the brain leaves plenty to be explored. At the lowest level, learning occurs through the strengthening of the synaptic connections between neurons. Of particular interest to us are neural networks and their evolution in response to stimuli. We present a Simulation Tool for Asynchronous Cortical Streams (STACS), which aims to address several key concerns with respect to learning in a neural network. Most importantly, we provide a framework for embodiment and feedback through interfacing with the external environment. Computationally, the simulator is designed from the ground up to run in parallel, taking into consideration the unique communication patterns of a highly connected, spiking neural network.