ITS Berkeley

Guardians of the Deep Fog: Failure-Resilient DNN Inference from Edge to Cloud

Yousefpour, Ashkan
Devic, Siddartha
Nguyen, Brian Q.
Kreidieh, Aboudy
Liao, Alan
Bayen, Alexandre M.
Jue, Jason P.
2019

Partitioning and distributing deep neural networks (DNNs) over physical nodes such as edge, fog, or cloud nodes, could enhance sensor fusion, and reduce bandwidth and inference latency. However, when a DNN is distributed over physical nodes, failure of the physical nodes causes the failure of the DNN units that are placed on these nodes. The performance of the inference task will be unpredictable, and most likely, poor, if the distributed DNN is not specifically designed and properly trained for failures. Motivated by this, we introduce deepFogGuard, a DNN architecture augmentation scheme...

Sensitivity Analysis and Relaxation of the Static Traffic Assignment Problem with Capacity Constraints

Cabannes, Theophile
Glista, Elizabeth
Dwarakanath, Kshama
Rao, Xu
Veeravalli, Tanya
Bayen, Alexandre M.
2019

This article introduces sensitivity analysis, reduction of the feasible set around the optimal solution, and LP and QP relaxations on convex, capacity-constrained network flow problems to evaluate the impact of a change in link capacity on the optimal flow allocation. This is done in the context of the static traffic assignment problem under capacity constraints [TAP-C], to understand the impact of traffic incidents on traffic flow in road networks, though the results also apply beyond transportation.The dual formulation of the convex [TAP-C] using gener-alized travel costs is exploited to...

Well-Posedness of Networked Scalar Semilinear Balance Laws Subject to Nonlinear Boundary Control Operators

Tang, Shu-Xia
Keimer, Alexander
Bayen, Alexandre M.
2019

Networked scalar semilinear balance laws are used as simplified macroscopic vehicular traffic models. The related initial boundary value problem is investigated, on a finite interval. The upstream boundary datum is determined by a nonlinear feedback control operator, representing the fact that traffic routing might be influenced in real time by the traffic information on the entire network. The main contribution of the present work lies in the appropriate design of nonlinear boundary control operators which meanwhile guarantee the well-posedness of the resultant systems. In detail, two...

A Study on Minimum Time Regulation of a Bounded Congested Road with Upstream Flow Control

Tang, Shu-Xia
Keimer, Alexander
Goatin, Paola
Bayen, Alexandre M.
2019

This article is motivated by the practical problem of controlling traffic flow by imposing restrictive boundary conditions. For a one-dimensional congested road segment, we study the minimum time control problem of how to control the upstream vehicular flow appropriately to regulate the downstream traffic into a desired (constant) free flow state in minimum time. We consider the Initial-Boundary Value Problem (IBVP) for a scalar nonlinear conservation law, associated to the Lighthill-Whitham-Richards (LWR) Partial Differential Equation (PDE), where the left boundary condition, also treated...

Inter-Level Cooperation in Hierarchical Reinforcement Learning

Rahman Kreidieh, Abdul
Berseth, Glen
Trabucco, Brandon
Parajuli, Samyak
Levine, Sergey
Bayen, Alexander M.
2019

Hierarchies of temporally decoupled policies present a promising approach for enabling structured exploration in complex long-term planning problems. To fully achieve this approach an end-to-end training paradigm is needed. However, training these multi-level policies has had limited success due to challenges arising from interactions between the goal-assigning and goal-achieving levels within a hierarchy. In this article, we consider the policy optimization process as a multi-agent process. This allows us to draw on connections between communication and cooperation in multi-agent RL, and...

Daily Data Assimilation of a Hydrologic Model Using the Ensemble Kalman Filter

Malek, Sami A.
Bayen, Alexandre M.
Glaser, Steven D.
2019

Accurate runoff forecasting is crucial for reservoir operators as it allows optimized water management, flood control and hydropower generation. Land surface models in mountainous regions depend on climatic inputs such as precipitation, temperature and solar radiation to model the water and energy dynamics and produce runoff as output. With the rapid development of cheap electronics applied in various systems, such as Wireless Sensor Networks (WSNs), satellite and airborne technologies, the prospect of practically measuring spatial Snow Water Equivalent in a dense temporal scale is...

Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design

Dennis, Michael
Jaques, Natasha
Vinitsky, Eugene
Bayen, Alexandre
Russell, Stuart
Critch, Andrew
Levine, Sergey
2020

A wide range of reinforcement learning (RL) problems --- including robustness, transfer learning, unsupervised RL, and emergent complexity --- require specifying a distribution of tasks or environments in which a policy will be trained. However, creating a useful distribution of environments is error prone, and takes a significant amount of developer time and effort. We propose Unsupervised Environment Design (UED) as an alternative paradigm, where developers provide environments with unknown parameters, and these parameters are used to automatically produce a distribution over valid,...

A Macroscopic Traffic Flow Model with Finite Buffers on Networks: Well-Posedness by Means of Hamilton-Jacobi Equations

Laurent-Brouty, Nicolas
Keimer, Alexander
Goatin, Paola
Bayen, Alexandre
2020

We introduce a model dealing with conservation laws on networks and coupled boundary conditions at the junctions. In particular, we introduce buffers of fixed arbitrary size and time dependent split ratios at the junctions , which represent how traffic is routed through the network, while guaranteeing spill-back phenomena at nodes. Having defined the dynamics at the level of conservation laws, we lift it up to the Hamilton-Jacobi (H-J) formulation and write boundary datum of incoming and outgoing junctions as functions of the queue sizes and vice-versa. The Hamilton-Jacobi formulation...

Long-Term Digital Device-Enabled Monitoring of Functional Status: Implications for Management of Persons with Alzheimer's Disease

Manley, Natalie A.
Bayen, Eleonore
Braley, Tamara L.
Merrilees, Jennifer
Clark, Amy M.
Zylstra, Bradley
Schaffer, Michael
Bayen, Alexandre
Possin, Katherine L.
2020

Introduction Informal caregiving is an essential element of health-care delivery. Little data describes how caregivers structure care recipients’ lives and impact their functional status. Methods We performed observational studies of community dwelling persons with dementia (PWD) to measure functional status by simultaneous assessment of physical activity (PA) and lifespace (LS). We present data from two caregiver/care-recipient dyads representing higher and average degrees of caregiver involvement. Results We acquired >42,800 (subject 1); >41,300 (subject 2) PA data points and >...

Learning Optimal Traffic Routing Behaviors Using Markovian Framework in Microscopic Simulation

Cabannes, T.
Li, J.
Wu, F.
Dong, H.
Bayen, A.M.
2020

This article applies the existing Markovian traffic assignment framework to novel traffic control strategies. In the Markovian traffic assignment framework, transition matrices are used to derive the traffic flow allocation. In contrast to the static traffic assignment, the framework only requires flow split ratio at every intersection, bypassing the need of computing path flow allocation. Consequently, compared to static traffic assignment, drivers’ routing behaviors can be modeled with fewer variables. As a result, it could be used to improve the efficiency of traffic management,...