arXiv subjects

Matthew A. Wright

Publications and source records attributed to Matthew A. Wright.

16 recordsLinked to original sources

arXiv2025

Robust electron counting for direct electron detectors with the Back-Propagation Counting method

Electron microscopy (EM) is a foundational tool for directly assessing the structure of materials. Recent advances in direct electron detectors have improved signal-to noise ratios via single-electron counting. However, accurately counting electrons at high fluence remains challenging. We developed a new method of electron counting for direct electron detectors, Back-Propagation Counting (BPC). BPC uses machine learning techniques designed for mathematical operations on large tensors but does not require large training datasets. In synthetic data, we show BPC is able to count multiple electron strikes per pixel and is robust to increasing occupancy. In experimental data, frames counted with BPC are shown to reconstruct diffraction peaks corresponding to individual nanoparticles with relatively higher intensity and produce images with improved contrast when compared to a standard counting method. Together, these results show that BPC excels in experiments where pixels see a high flux of electron irradiation such as in situ TEM movies and diffraction.

physics.ins-det↗

arXiv2022

Representing Long-Range Context for Graph Neural Networks with Global Attention

Graph neural networks are powerful architectures for structured datasets. However, current methods struggle to represent long-range dependencies. Scaling the depth or width of GNNs is insufficient to broaden receptive fields as larger GNNs encounter optimization instabilities such as vanishing gradients and representation oversmoothing, while pooling-based approaches have yet to become as universally useful as in computer vision. In this work, we propose the use of Transformer-based self-attention to learn long-range pairwise relationships, with a novel "readout" mechanism to obtain a global graph embedding. Inspired by recent computer vision results that find position-invariant attention performant in learning long-range relationships, our method, which we call GraphTrans, applies a permutation-invariant Transformer module after a standard GNN module. This simple architecture leads to state-of-the-art results on several graph classification tasks, outperforming methods that explicitly encode graph structure. Our results suggest that purely-learning-based approaches without graph structure may be suitable for learning high-level, long-range relationships on graphs. Code for GraphTrans is available at https://github.com/ucbrise/graphtrans.

cs.LG↗

arXiv2021

Transformers are Deep Infinite-Dimensional Non-Mercer Binary Kernel Machines

Despite their ubiquity in core AI fields like natural language processing, the mechanics of deep attention-based neural networks like the Transformer model are not fully understood. In this article, we present a new perspective towards understanding how Transformers work. In particular, we show that the "dot-product attention" that is the core of the Transformer's operation can be characterized as a kernel learning method on a pair of Banach spaces. In particular, the Transformer's kernel is characterized as having an infinite feature dimension. Along the way we consider an extension of the standard kernel learning problem to a binary setting, where data come from two input domains and a response is defined for every cross-domain pair. We prove a new representer theorem for these binary kernel machines with non-Mercer (indefinite, asymmetric) kernels (implying that the functions learned are elements of reproducing kernel Banach spaces rather than Hilbert spaces), and also prove a new universal approximation theorem showing that the Transformer calculation can learn any binary non-Mercer reproducing kernel Banach space pair. We experiment with new kernels in Transformers, and obtain results that suggest the infinite dimensionality of the standard Transformer kernel is partially responsible for its performance. This paper's results provide a new theoretical understanding of a very important but poorly understood model in modern machine~learning.

cs.LG↗

arXiv2021

Contingencies from Observations: Tractable Contingency Planning with Learned Behavior Models

Humans have a remarkable ability to make decisions by accurately reasoning about future events, including the future behaviors and states of mind of other agents. Consider driving a car through a busy intersection: it is necessary to reason about the physics of the vehicle, the intentions of other drivers, and their beliefs about your own intentions. If you signal a turn, another driver might yield to you, or if you enter the passing lane, another driver might decelerate to give you room to merge in front. Competent drivers must plan how they can safely react to a variety of potential future behaviors of other agents before they make their next move. This requires contingency planning: explicitly planning a set of conditional actions that depend on the stochastic outcome of future events. In this work, we develop a general-purpose contingency planner that is learned end-to-end using high-dimensional scene observations and low-dimensional behavioral observations. We use a conditional autoregressive flow model to create a compact contingency planning space, and show how this model can tractably learn contingencies from behavioral observations. We developed a closed-loop control benchmark of realistic multi-agent scenarios in a driving simulator (CARLA), on which we compare our method to various noncontingent methods that reason about multi-agent future behavior, including several state-of-the-art deep learning-based planning approaches. We illustrate that these noncontingent planning methods fundamentally fail on this benchmark, and find that our deep contingency planning method achieves significantly superior performance. Code to run our benchmark and reproduce our results is available at https://sites.google.com/view/contingency-planning

cs.RO↗

arXiv2021

Pylot: A Modular Platform for Exploring Latency-Accuracy Tradeoffs in Autonomous Vehicles

We present Pylot, a platform for autonomous vehicle (AV) research and development, built with the goal to allow researchers to study the effects of the latency and accuracy of their models and algorithms on the end-to-end driving behavior of an AV. This is achieved through a modular structure enabled by our high-performance dataflow system that represents AV software pipeline components (object detectors, motion planners, etc.) as a dataflow graph of operators which communicate on data streams using timestamped messages. Pylot readily interfaces with popular AV simulators like CARLA, and is easily deployable to real-world vehicles with minimal code changes. To reduce the burden of developing an entire pipeline for evaluating a single component, Pylot provides several state-of-the-art reference implementations for the various components of an AV pipeline. Using these reference implementations, a Pylot-based AV pipeline is able to drive a real vehicle, and attains a high score on the CARLA Autonomous Driving Challenge. We also present several case studies enabled by Pylot, including evidence of a need for context-dependent components, and per-component time allocation. Pylot is open source, with the code available at https://github.com/erdos-project/pylot.

cs.RO↗

arXiv2019

Macroscopic Modeling, Calibration, and Simulation of Managed Lane-Freeway Networks, Part II: Network-scale Calibration and Case Studies

In Part I of this paper series, several macroscopic traffic model elements for mathematically describing freeway networks equipped with managed lane facilities were proposed. These modeling techniques seek to capture at the macroscopic the complex phenomena that occur on managed lane-freeway networks, where two parallel traffic flows interact with each other both in the physical sense (how and where cars flow between the two lane groups) and the physiological sense (how driving behaviors are changed by being adjacent to a quantitatively and qualitatively different traffic flow). The local descriptions we developed in Part I are not the only modeling complexity introduced in managed lane-freeway networks. The complex topologies mean that network-scale modeling of a freeway corridor is increased in complexity as well. The already-difficult model calibration problem for a dynamic model of a freeway becomes more complex when the freeway becomes, in effect, two interrelating flow streams. In the present paper, we present an iterative-learning-based approach to calibrating our model's physical and driver-behavioral parameters. We consider the common situation where a complex traffic model needs to be calibrated to recreate real-world baseline traffic behavior, such that counterfactuals can be generated by training purposes. Our method is used to identify traditional freeway parameters as well as the proposed parameters that describe managed lane-freeway-network-specific behaviors. We validate our model and calibration methodology with case studies of simulations of two managed lane-equipped California freeways.

eess.SY↗

arXiv2019

Neural-Attention-Based Deep Learning Architectures for Modeling Traffic Dynamics on Lane Graphs

Deep neural networks can be powerful tools, but require careful application-specific design to ensure that the most informative relationships in the data are learnable. In this paper, we apply deep neural networks to the nonlinear spatiotemporal physics problem of vehicle traffic dynamics. We consider problems of estimating macroscopic quantities (e.g., the queue at an intersection) at a lane level. First-principles modeling at the lane scale has been a challenge due to complexities in modeling social behaviors like lane changes, and those behaviors' resultant macro-scale effects. Following domain knowledge that upstream/downstream lanes and neighboring lanes affect each others' traffic flows in distinct ways, we apply a form of neural attention that allows the neural network layers to aggregate information from different lanes in different manners. Using a microscopic traffic simulator as a testbed, we obtain results showing that an attentional neural network model can use information from nearby lanes to improve predictions, and, that explicitly encoding the lane-to-lane relationship types significantly improves performance. We also demonstrate the transfer of our learned neural network to a more complex road network, discuss how its performance degradation may be attributable to new traffic behaviors induced by increased topological complexity, and motivate learning dynamics models from many road network topologies.

cs.LG↗

arXiv2019

Generic second-order macroscopic traffic node model for general multi-input multi-output road junctions via a dynamic system approach

This paper addresses an open problem in traffic modeling: the second-order macroscopic node problem. A second-order macroscopic traffic model, in contrast to a first-order model, allows for variation of driving behavior across subpopulations of vehicles in the flow. The second-order models are thus more descriptive (e.g., they have been used to model variable mixtures of behaviorally-different traffic, like car/truck traffic, autonomous/human-driven traffic, etc.), but are much more complex. The second-order node problem is a particularly complex problem, as it requires the resolution of discontinuities in traffic density and mixture characteristics, and solving of throughflows for arbitrary numbers of input and output roads to a node (in other words, this is an arbitrary-dimensional Riemann problem with two conserved quantities). In this paper, we extend the well-known "Generic Class of Node Model" constraints to the second order and present a simple solution algorithm to the second-order node problem. Our solution makes use of a recently-introduced dynamic system characterization of the first-order node model problem, which gives insight and intuition as to the continuous-time dynamics implicit in node models. We further argue that the common "supply and demand" construction of node models that decouples them from link models is not suitable to the second-order node problem. Our second-order node model and solution method have immediate applications in allowing modeling of behaviorally-complex traffic flows of contemporary interest (like partially-autonomous-vehicle flows) in arbitrary road networks.

eess.SY↗

arXiv2019

Macroscopic Modeling, Calibration, and Simulation of Managed Lane-Freeway Networks, Part I: Topological and Phenomenological Modeling

To help mitigate road congestion caused by the unrelenting growth of traffic demand, many transit authorities have implemented managed lane policies. Managed lanes typically run parallel to a freeway's standard, general-purpose (GP) lanes, but are restricted to certain types of vehicles. It was originally thought that managed lanes would improve the use of existing infrastructure through incentivization of demand-management behaviors like carpooling, but implementations have often been characterized by unpredicted phenomena that is often to detrimental system performance. This paper presents several macroscopic traffic modeling tools we have used for study of freeways equipped with managed lanes, or "managed lane-freeway networks." The proposed framework is based on the widely-used first-order kinematic wave theory. In this model, the GP and the managed lanes are modeled as parallel links connected by nodes, where certain type of traffic may switch between GP and managed lane links. Two types of managed lane topologies are considered: full-access, where vehicles can switch between the GP and the managed lanes anywhere; and separated, where such switching is allowed only at certain locations called gates. We also describe methods to incorporate in three phenomena into our model that are particular to managed lane-freeway networks. The inertia effect reflects drivers' inclination to stay in their lane as long as possible and switch only if this would obviously improve their travel condition. The friction effect reflects the empirically-observed driver fear of moving fast in a managed lane while traffic in the adjacent GP lanes moves slowly due to congestion. The smoothing effect describes how managed lanes can increase throughput at bottlenecks by reducing lane changes. We present simple models for each of these phenomena that fit within the general macroscopic theory.

eess.SY↗

arXiv2019

Attentional Policies for Cross-Context Multi-Agent Reinforcement Learning

Many potential applications of reinforcement learning in the real world involve interacting with other agents whose numbers vary over time. We propose new neural policy architectures for these multi-agent problems. In contrast to other methods of training an individual, discrete policy for each agent and then enforcing cooperation through some additional inter-policy mechanism, we follow the spirit of recent work on the power of relational inductive biases in deep networks by learning multi-agent relationships at the policy level via an attentional architecture. In our method, all agents share the same policy, but independently apply it in their own context to aggregate the other agents' state information when selecting their next action. The structure of our architectures allow them to be applied on environments with varying numbers of agents. We demonstrate our architecture on a benchmark multi-agent autonomous vehicle coordination problem, obtaining superior results to a full-knowledge, fully-centralized reference solution, and significantly outperforming it when scaling to large numbers of agents.

cs.LG↗

arXiv2019

A Framework for Robust Assimilation of Potentially Malign Third-Party Data, and its Statistical Meaning

This paper presents a model-based method for fusing data from multiple sensors with a hypothesis-test-based component for rejecting potentially faulty or otherwise malign data. Our framework is based on an extension of the classic particle filter algorithm for real-time state estimation of uncertain systems with nonlinear dynamics with partial and noisy observations. This extension, based on classical statistical theories, utilizes statistical tests against the system's observation model. We discuss the application of the two major statistical testing frameworks, Fisherian significance testing and Neyman-Pearsonian hypothesis testing, to the Monte Carlo and sensor fusion settings. The Monte Carlo Neyman-Pearson test we develop is useful when one has a reliable model of faulty data, while the Fisher one is applicable when one may not have a model of faults, which may occur when dealing with third-party data, like GNSS data of transportation system users. These statistical tests can be combined with a particle filter to obtain a Monte Carlo state estimation scheme that is robust to faulty or outlier data. We present a synthetic freeway traffic state estimation problem where the filters are able to reject simulated faulty GNSS measurements. The fault-model-free Fisher filter, while underperforming the Neyman-Pearson one when the latter has an accurate fault model, outperforms it when the assumed fault model is incorrect.

eess.SY↗

arXiv2018

A Dynamic-System-Based Approach to Modeling Driver Movements Across General-Purpose/Managed Lane Interfaces

To help mitigate road congestion caused by the unrelenting growth of traffic demand, many transportation authorities have implemented managed lane policies, which restrict certain freeway lanes to certain types of vehicles. It was originally thought that managed lanes would improve the use of existing infrastructure through demand-management behaviors like carpooling, but implementations have often been characterized by unpredicted phenomena that are sometimes detrimental to system performance. The development of traffic models that can capture these sorts of behaviors is a key step for helping managed lanes deliver on their promised gains. Towards this goal, this paper presents an approach for solving for driver behavior of entering and exiting managed lanes at the macroscopic (i.e., fluid approximation of traffic) scale. Our method is inspired by recent work in extending a dynamic-system-based modeling framework from traffic behaviors on individual roads, to models at junctions, and can be considered a further extension of this dynamic-system paradigm to the route/lane choice problem. Unlike traditional route choice models that are often based on discrete-choice methods and often rely on computing and comparing drivers' estimated travel times from taking different routes, our method is agnostic to the particular choice of physical traffic model and is suited specifically towards making decisions at these interfaces using only local information. These features make it a natural drop-in component to extend existing dynamic traffic modeling methods.

eess.SY↗

arXiv2018

A Community-Developed Open-Source Computational Ecosystem for Big Neuro Data

Big imaging data is becoming more prominent in brain sciences across spatiotemporal scales and phylogenies. We have developed a computational ecosystem that enables storage, visualization, and analysis of these data in the cloud, thusfar spanning 20+ publications and 100+ terabytes including nanoscale ultrastructure, microscale synaptogenetic diversity, and mesoscale whole brain connectivity, making NeuroData the largest and most diverse open repository of brain data.

q-bio.NC↗

arXiv2017

Particle-Filter-Enabled Real-Time Sensor Fault Detection Without a Model of Faults

We are experiencing an explosion in the amount of sensors measuring our activities and the world around us. These sensors are spread throughout the built environment and can help us perform state estimation and control of related systems, but they are often built and/or maintained by third parties or system users. As a result, by outsourcing system measurement to third parties, the controller must accept their measurements without being able to directly verify the sensors' correct operation. Instead, detection and rejection of measurements from faulty sensors must be done with the raw data only. Towards this goal, we present a method of detecting possibly faulty behavior of sensors. The method does not require that the control designer have any model of faulty sensor behavior. As we discuss, it turns out that the widely-used particle filter state estimation algorithm provides the ingredients necessary for a hypothesis test against all ranges of correct operating behavior, obviating the need for a fault model to compare measurements. We demonstrate the applicability of our method by demonstrating its ability to reject faulty measurements and improve state estimation accuracy in a nonlinear vehicle traffic model without information of generated faulty measurements' characteristics. In our test, we correctly identify nearly 90% of measurements as faulty or non-faulty without having any fault model. This leads to only a 3% increase in state estimation error over a theoretical 100%-accurate fault detector.

eess.SY↗

arXiv2017

On node models for high-dimensional road networks

Macroscopic traffic models are necessary for simulation and study of traffic's complex macro-scale dynamics, and are often used by practitioners for road network planning, integrated corridor management, and other applications. These models have two parts: a link model, which describes traffic flow behavior on individual roads, and a node model, which describes behavior at road junctions. As the road networks under study become larger and more complex --- nowadays often including arterial networks --- the node model becomes more important. This paper focuses on the first order node model and has two main contributions. First, we formalize the multi-commodity flow distribution at a junction as an optimization problem with all the necessary constraints. Most interesting here is the formalization of input flow priorities. Then, we discuss a very common "conservation of turning fractions" or "first-in-first-out" (FIFO) constraint, and how it often produces unrealistic spillback. This spillback occurs when, at a diverge, a queue develops for a movement that only a few lanes service, but FIFO requires that all lanes experience spillback from this queue. As we show, avoiding this unrealistic spillback while retaining FIFO in the node model requires complicated network topologies. Our second contribution is a "partial FIFO" mechanism that avoids this unrealistic spillback, and a node model and solution algorithm that incorporates this mechanism. The partial FIFO mechanism is parameterized through intervals that describe how individual movements influence each other, can be intuitively described from physical lane geometry and turning movement rules, and allows tuning to describe a link as having anything between full FIFO and no FIFO. Excepting the FIFO constraint, the present node model also fits within the well-established "general class of first-order node models" for multi-commodity flows.

eess.SY↗

arXiv2016

A dynamic system characterization of road network node models

The propagation of traffic congestion along roads is a commonplace nonlinear phenomenon. When many roads are connected in a network, congestion can spill from one road to others as drivers queue to enter a congested road, creating further nonlinearities in the network dynamics. This paper considers the node model problem, which refers to methods for solving for cross-flows when roads meet at a junction. We present a simple hybrid dynamic system that, given a macroscopic snapshot of the roads entering and exiting a node, intuitively models the node's throughflows over time. This dynamic system produces solutions to the node model problem that are equal to those produced by many popular node models without intuitive physical meanings. We also show how the earlier node models can be rederived as executions of our dynamic system. The intuitive physical description supplied by our system provides a base for control of the road junction system dynamics, as well as the emergent network dynamics.

eess.SY↗