arXiv subjects

Gunjan Verma

Publications and source records attributed to Gunjan Verma.

35 records · Page 2Linked to original sources

arXiv2022

Link Scheduling using Graph Neural Networks

Efficient scheduling of transmissions is a key problem in wireless networks. The main challenge stems from the fact that optimal link scheduling involves solving a maximum weighted independent set (MWIS) problem, which is known to be NP-hard. In practical schedulers, centralized and distributed greedy heuristics are commonly used to approximately solve the MWIS problem. However, most of these greedy heuristics ignore important topological information of the wireless network. To overcome this limitation, we propose fast heuristics based on graph convolutional networks (GCNs) that can be implemented in centralized and distributed manners. Our centralized heuristic is based on tree search guided by a GCN and 1-step rollout. In our distributed MWIS solver, a GCN generates topology-aware node embeddings that are combined with per-link utilities before invoking a distributed greedy solver. Moreover, a novel reinforcement learning scheme is developed to train the GCN in a non-differentiable pipeline. Test results on medium-sized wireless networks show that our centralized heuristic can reach a near-optimal solution quickly, and our distributed heuristic based on a shallow GCN can reduce by nearly half the suboptimality gap of the distributed greedy solver with minimal increase in complexity. The proposed schedulers also exhibit good generalizability across graph and weight distributions.

eess.SP↗

arXiv2022

Adversarial Examples in Constrained Domains

Machine learning algorithms have been shown to be vulnerable to adversarial manipulation through systematic modification of inputs (e.g., adversarial examples) in domains such as image recognition. Under the default threat model, the adversary exploits the unconstrained nature of images; each feature (pixel) is fully under control of the adversary. However, it is not clear how these attacks translate to constrained domains that limit which and how features can be modified by the adversary (e.g., network intrusion detection). In this paper, we explore whether constrained domains are less vulnerable than unconstrained domains to adversarial example generation algorithms. We create an algorithm for generating adversarial sketches: targeted universal perturbation vectors which encode feature saliency within the envelope of domain constraints. To assess how these algorithms perform, we evaluate them in constrained (e.g., network intrusion detection) and unconstrained (e.g., image recognition) domains. The results demonstrate that our approaches generate misclassification rates in constrained domains that were comparable to those of unconstrained domains (greater than 95%). Our investigation shows that the narrow attack surface exposed by constrained domains is still sufficiently large to craft successful adversarial examples; and thus, constraints do not appear to make a domain robust. Indeed, with as little as five randomly selected features, one can still generate adversarial examples.

cs.CR↗

arXiv2022

Delay-Oriented Distributed Scheduling Using Graph Neural Networks

In wireless multi-hop networks, delay is an important metric for many applications. However, the max-weight scheduling algorithms in the literature typically focus on instantaneous optimality, in which the schedule is selected by solving a maximum weighted independent set (MWIS) problem on the interference graph at each time slot. These myopic policies perform poorly in delay-oriented scheduling, in which the dependency between the current backlogs of the network and the schedule of the previous time slot needs to be considered. To address this issue, we propose a delay-oriented distributed scheduler based on graph convolutional networks (GCNs). In a nutshell, a trainable GCN module generates node embeddings that capture the network topology as well as multi-step lookahead backlogs, before calling a distributed greedy MWIS solver. In small- to medium-sized wireless networks with heterogeneous transmit power, where a few central links have many interfering neighbors, our proposed distributed scheduler can outperform the myopic schedulers based on greedy and instantaneously optimal MWIS solvers, with good generalizability across graph models and minimal increase in communication complexity.

eess.SP↗

arXiv2021

ML-aided power allocation for Tactical MIMO

We study the problem of optimal power allocation in single-hop multi-antenna ad-hoc wireless networks. A standard technique to solve this problem involves optimizing a tri-convex function under power constraints using a block-coordinate-descent based iterative algorithm. This approach, termed WMMSE, tends to be computationally complex and time consuming. Several learning-based approaches have been proposed to speed up the power allocation process. A recent work, UWMMSE, learns an affine transformation of a WMMSE parameter in an unfolded structure to accelerate convergence. In spite of achieving promising results, its application is limited to single-antenna wireless networks. In this work, we present a UWMMSE framework for power allocation in (multiple-input multiple-output) MIMO interference networks. A major advantage of this method lies in its use of low-complexity learnable systems in which the number of parameters scales linearly with respect to the hidden layer size of embedded neural architectures and the product of the number of transmitter and receiver antennas only, fully independent of the number of transceivers in the network. We illustrate the superiority of our method through an empirical study of our approach in comparison to WMMSE and also analyze its robustness to changes in channel conditions and network size.

cs.IT↗

arXiv2021

Unfolding WMMSE using Graph Neural Networks for Efficient Power Allocation

We study the problem of optimal power allocation in a single-hop ad hoc wireless network. In solving this problem, we depart from classical purely model-based approaches and propose a hybrid method that retains key modeling elements in conjunction with data-driven components. More precisely, we put forth a neural network architecture inspired by the algorithmic unfolding of the iterative weighted minimum mean squared error (WMMSE) method, that we denote by unfolded WMMSE (UWMMSE). The learnable weights within UWMMSE are parameterized using graph neural networks (GNNs), where the time-varying underlying graphs are given by the fading interference coefficients in the wireless network. These GNNs are trained through a gradient descent approach based on multiple instances of the power allocation problem. We show that the proposed architecture is permutation equivariant, thus facilitating generalizability across network topologies. Comprehensive numerical experiments illustrate the performance attained by UWMMSE along with its robustness to hyper-parameter selection and generalizability to unseen scenarios such as different network densities and network sizes.

eess.SP↗

arXiv2021

Distributed Scheduling using Graph Neural Networks

A fundamental problem in the design of wireless networks is to efficiently schedule transmission in a distributed manner. The main challenge stems from the fact that optimal link scheduling involves solving a maximum weighted independent set (MWIS) problem, which is NP-hard. For practical link scheduling schemes, distributed greedy approaches are commonly used to approximate the solution of the MWIS problem. However, these greedy schemes mostly ignore important topological information of the wireless networks. To overcome this limitation, we propose a distributed MWIS solver based on graph convolutional networks (GCNs). In a nutshell, a trainable GCN module learns topology-aware node embeddings that are combined with the network weights before calling a greedy solver. In small- to middle-sized wireless networks with tens of links, even a shallow GCN-based MWIS scheduler can leverage the topological information of the graph to reduce in half the suboptimality gap of the distributed greedy solver with good generalizability across graphs and minimal increase in complexity.

eess.SP↗

arXiv2020

Adaptive Contention Window Design using Deep Q-learning

We study the problem of adaptive contention window (CW) design for random-access wireless networks. More precisely, our goal is to design an intelligent node that can dynamically adapt its minimum CW (MCW) parameter to maximize a network-level utility knowing neither the MCWs of other nodes nor how these change over time. To achieve this goal, we adopt a reinforcement learning (RL) framework where we circumvent the lack of system knowledge with local channel observations and we reward actions that lead to high utilities. To efficiently learn these preferred actions, we follow a deep Q-learning approach, where the Q-value function is parametrized using a multi-layer perception. In particular, we implement a rainbow agent, which incorporates several empirical improvements over the basic deep Q-network. Numerical experiments based on the NS3 simulator reveal that the proposed RL agent performs close to optimal and markedly improves upon existing learning and non-learning based alternatives.

eess.SP↗

arXiv2020

Efficient power allocation using graph neural networks and deep algorithm unfolding

We study the problem of optimal power allocation in a single-hop ad hoc wireless network. In solving this problem, we propose a hybrid neural architecture inspired by the algorithmic unfolding of the iterative weighted minimum mean squared error (WMMSE) method, that we denote as unfolded WMMSE (UWMMSE). The learnable weights within UWMMSE are parameterized using graph neural networks (GNNs), where the time-varying underlying graphs are given by the fading interference coefficients in the wireless network. These GNNs are trained through a gradient descent approach based on multiple instances of the power allocation problem. Once trained, UWMMSE achieves performance comparable to that of WMMSE while significantly reducing the computational complexity. This phenomenon is illustrated through numerical experiments along with the robustness and generalization to wireless networks of different densities and sizes.

eess.SP↗

arXiv2020

Canceling the cavity length induced phase noise in an optical ring cavity for phase shift measurement and spin squeezing

We demonstrate a new method of light phase shift measurement using a high-finesse optical ring cavity which exhibits reduced phase noise due to cavity length fluctuations. Two laser beams with a frequency difference of one cavity free spectral range are simultaneously resonant with the cavity, demonstrating noise correlations in the error signals due to the common-mode cavity length fluctuations. The differential error signal shows a 30 dB reduction in cavity noise down to the noise floor in a frequency range up to half the cavity linewidth ($δν/2 \simeq 30$ kHz). Various noise sources are analyzed and their contributions to the noise floor are evaluated. Additionally, we apply this noise-reduced phase shift measurement scheme in a simulated spin-squeezing experiment where we have achieved a factor of 40 improvement in phase sensitivity with a phase resolution of 0.7 mrad, which may remove one important barrier against attaining highly spin-squeezed states. The demonstrated method is the first reported measurement using an optical ring cavity and two independent beams, a flexible situation. This method can find direct application to non-destructive measurements in quantum systems, such as for the generation of spin-squeezed states in atom interferometers and atomic clocks.

physics.atom-ph↗

arXiv2020

Rapid Top-Down Synthesis of Large-Scale IoT Networks

Advances in optimization and constraint satisfaction techniques, together with the availability of elastic computing resources, have spurred interest in large-scale network verification and synthesis. Motivated by this, we consider the top-down synthesis of ad-hoc IoT networks for disaster response and search and rescue operations. This synthesis problem must satisfy complex and competing constraints: sensor coverage, line-of-sight visibility, and network connectivity. The central challenge in our synthesis problem is quickly scaling to large regions while producing cost-effective solutions. We explore two qualitatively different representations of the synthesis problems satisfiability modulo convex optimization (SMC), and mixed-integer linear programming (MILP). The former is more expressive, for our problem, than the latter, but is less well-suited for solving optimization problems like ours. We show how to express our network synthesis in these frameworks, and, to scale to problem sizes beyond what these frameworks are capable of, develop a hierarchical synthesis technique that independently synthesizes networks in sub-regions of the deployment area, then combines these. We find that, while MILP outperforms SMC in some settings for smaller problem sizes, the fact that SMC's expressivity matches our problem ensures that it uniformly generates better quality solutions at larger problem sizes.

cs.NI↗

arXiv2019

Information Flow Optimization in Inference Networks

The problem of maximizing the information flow through a sensor network tasked with an inference objective at the fusion center is considered. The sensor nodes take observations, compress and send them to the fusion center through a network of relays. The network imposes capacity constraints on the rate of transmission in each connection and flow conservation constraints. It is shown that this rate-constrained inference problem can be cast as a Network Utility Maximization problem by suitably defining the utility functions for each sensor, and can be solved using existing techniques. Two practical settings are analyzed: multi-terminal parameter estimation and binary hypothesis testing. It is verified via simulations that using the proposed formulation gives better inference performance than the Max-Flow solution that simply maximizes the total bit-rate to the fusion center.

math.OC↗

arXiv2019

Attribution-driven Causal Analysis for Detection of Adversarial Examples

Attribution methods have been developed to explain the decision of a machine learning model on a given input. We use the Integrated Gradient method for finding attributions to define the causal neighborhood of an input by incrementally masking high attribution features. We study the robustness of machine learning models on benign and adversarial inputs in this neighborhood. Our study indicates that benign inputs are robust to the masking of high attribution features but adversarial inputs generated by the state-of-the-art adversarial attack methods such as DeepFool, FGSM, CW and PGD, are not robust to such masking. Further, our study demonstrates that this concentration of high-attribution features responsible for the incorrect decision is more pronounced in physically realizable adversarial examples. This difference in attribution of benign and adversarial inputs can be used to detect adversarial examples. Such a defense approach is independent of training data and attack method, and we demonstrate its effectiveness on digital and physically realizable perturbations.

cs.LG↗

arXiv2018

A simple atomic beam oven with a metal thermal break

We report the design and construction of a simple, easy to machine high-temperature oven for generating an atomic beam in laser cooling experiments. This design eliminates the problem of thermal isolation of the oven region from the rest of the vacuum system without using a glass or ceramic thermal break. This design simplifies the construction and operation of high-temperature ovens for elements having low vapor pressure. We demonstrate the functionality of such a source for Strontium (Sr) atoms. We generate a high flux of Sr atoms for use in laser cooling and trapping experiments. The optimization of the design of the metal thermal break is done using a finite element analysis.

physics.atom-ph↗

arXiv2017

Dark solitons generation and their instability dynamics in two dimensional condensates

We analyze numerically the formation and the subsequent dynamics of two-dimensional matter wave dark solitons in a Thomas-Fermi rubidium condensate using various techniques. An initially imprinted sharp phase gradient leads to the dynamical formation of a stationary soliton as well as very shallow grey solitons, whereas a smooth gradient only creates grey solitons. The depth and hence, the velocity of the soliton is provided by the spatial width of the phase gradient, and it also strongly influences the snake-instability dynamics of the two dimensional solitons. The vortex dipoles stemming from the unstable soliton exhibit rich dynamics. Notably, the annihilation of a vortex dipole via a transient dark lump or a vortexonium state, the exchange of vortices between either a pair of vortex dipoles or a vortex dipole and a single vortex, and so on. For sufficiently large width of the initial phase gradient, the solitons may decay directly into vortexoniums instead of vortex pairs, and also the decay rate is augmented. Later, we discuss alternative techniques to generate dark solitons, which involve a Gaussian potential barrier and time-dependent interactions, both linear and periodic. The properties of the solitons can be controlled by tuning the amplitude or the width of the potential barrier. In the linear case, the number of solitons and their depths are determined by the quench time of the interactions. For the periodic modulation, a transient soliton lattice emerges with its periodicity depending on the modulation frequency, through a wave number selection governed by the local Bogoliubov spectrum. Interestingly, for sufficiently low barrier potential, both Faraday pattern and soliton lattice coexist. The snake instability dynamics of the soliton lattice is characteristically modified if the Faraday pattern is present.

cond-mat.quant-gas↗

arXiv2017

Non-exponential decoherence and subdiffusion in atom-optics kicked rotor

Quantum systems lose coherence upon interaction with the environment and tend towards classical states. Quantum coherence is known to exponentially decay in time so that macroscopic quantum superpositions are generally unsustainable. In this work, slower than exponential decay of coherences is experimentally realized in an atom-optics kicked rotor system subjected to non-stationary Lévy noise in the applied kick sequence. The slower coherence decay manifests in the form of quantum subdiffusion that can be controlled through the Lévy exponent. The experimental results are in good agreement with the analytical estimates and numerical simulations for the mean energy growth and momentum profiles of atom-optics kicked rotor.

quant-ph↗

arXiv2017

A compact atomic beam based system for Doppler-free laser spectroscopy of Strontium atoms

We report the construction of a simple, light weight and compact atomic beam spectroscopy cell for Strontium atoms. The cell is built using glass blowing technique and includes a simple Titanium Sublimation Pump for active pumping of the residual and background gases to maintain ultra-high vacuum. Commercially available and electrically heated dispenser source is used to generate the beam of Sr atoms. We perform spectroscopy on the $5s^2\ ^1S_0\longrightarrow 5s\ 5p\ ^1P_1$ transition to obtain a well resolved Doppler free spectroscopic signal for frequency stabilization of the laser source. This design can be easily extended to other alkali and alkaline earth metals.

physics.atom-ph↗

arXiv2014

Bose-Einstein Condensation in an electro-pneumatically transformed quadrupole-Ioffe magnetic trap

We report a novel approach for preparing a Bose-Einstein condensate (BEC) of $^{87}$Rb atoms using electro-pneumatically driven transfer of atoms into a Quadrupole-Ioffe magnetic trap (QUIC Trap). More than 5$\times$$10^{8}$ atoms from a Magneto-optical trap are loaded into a spherical quadrupole trap and then these atoms are transferred into an Ioffe trap by moving the Ioffe coil towards the center of the quadrupole coil, thereby, changing the distance between quadrupole trap center and the Ioffe coil. The transfer efficiency is more than 80 \%. This approach is different from a conventional approach of loading the atoms into a QUIC trap wherein the spherical quadrupole trap is transformed into a QUIC trap by changing the currents in the quadrupole and the Ioffe coils. The phase space density is then increased by forced rf evaporative cooling to achieve the Bose-Einstein condensation having more than $10^{5}$ atoms.

physics.atom-ph↗