arXiv subjects

Runxin He

Publications and source records attributed to Runxin He.

7 recordsLinked to original sources

arXiv2023

ONNXExplainer: an ONNX Based Generic Framework to Explain Neural Networks Using Shapley Values

Understanding why a neural network model makes certain decisions can be as important as the inference performance. Various methods have been proposed to help practitioners explain the prediction of a neural network model, of which Shapley values are most popular. SHAP package is a leading implementation of Shapley values to explain neural networks implemented in TensorFlow or PyTorch but lacks cross-platform support, one-shot deployment and is highly inefficient. To address these problems, we present the ONNXExplainer, which is a generic framework to explain neural networks using Shapley values in the ONNX ecosystem. In ONNXExplainer, we develop its own automatic differentiation and optimization approach, which not only enables One-Shot Deployment of neural networks inference and explanations, but also significantly improves the efficiency to compute explanation with less memory consumption. For fair comparison purposes, we also implement the same optimization in TensorFlow and PyTorch and measure its performance against the current state of the art open-source counterpart, SHAP. Extensive benchmarks demonstrate that the proposed optimization approach improves the explanation latency of VGG19, ResNet50, DenseNet201, and EfficientNetB0 by as much as 500%.

cs.LG↗

arXiv2020

DL-IAPS and PJSO: A Path/Speed Decoupled Trajectory Optimization and its Application in Autonomous Driving

This paper presents a free space trajectory optimization algorithm of autonomous driving vehicle, which decouples the collision-free trajectory planning problem into a Dual-Loop Iterative Anchoring Path Smoothing (DL-IAPS) and a Piece-wise Jerk Speed Optimization (PJSO). The work leads to remarkable driving performance improvements including more precise collision avoidance, higher control feasibility and better driving comfort, as those are often hard to realize in other existing path/speed decoupled trajectory optimization methods. Our algorithm's efficiency, robustness and adaptiveness to complex driving scenarios have been validated by both simulations and real on-road tests.

cs.RO↗

arXiv2020

TDR-OBCA: A Reliable Planner for Autonomous Driving in Free-Space Environment

This paper presents an optimization-based collision avoidance trajectory generation method for autonomous driving in free-space environments, with enhanced robustness, driving comfort and efficiency. Starting from the hybrid optimization-based framework, we introduces two warm start methods, temporal and dual variable warm starts, to improve the efficiency. We also reformulate the problem to improve the robustness and efficiency. We name this new algorithm TDR-OBCA. With these changes, compared with original hybrid optimization we achieve a 96.67% failure rate decrease with respect to initial conditions, 13.53% increase in driving comforts and 3.33% to 44.82% increase in planner efficiency as obstacles number scales. We validate our results in hundreds of simulation scenarios and hundreds of hours of public road tests in both U.S. and China. Our source code is available at https://github.com/ApolloAuto/apollo.

cs.RO↗

arXiv2017

Comfort-Aware Building Climate Control Using Distributed-Parameter Models

Controlling Heating, Ventilation and Air Conditioning (HVAC) system to maintain occupant's indoor thermal comfort is important to energy-efficient buildings and the development of smart cities. In this paper, we formulate a model predictive controller (MPC) to make optimal control strategies to HVAC in order to maintain occupant's comfort by predicted mean vote index, and then present a whole system with an estimator to indoor climate and apartment's geometric information based on only thermostats. In order to have accurate spatial resolution and make the HVAC system focus on only a zoned area around the occupant, a convection-diffusion Computer Fluid Dynamics (CFD) model is used to describe the indoor air flow and temperature distribution. The MPC system generates corresponding PDE-constrained optimization problems, and we solve them by obtain the gradients of cost functions with respect to problems' variables with the help of CFD model's adjoint equations. We evaluate the performance of our method using simulations of a real apartment in the St.\ Louis area. Our results show our MPC system's energy efficiency and the potential for its application in real-time operation of high-performance buildings.

math.OC↗

arXiv2017

Numerical Synthesis of Pontryagin Optimal Control Minimizers Using Sampling-Based Methods

Optimal control remains as one of the most versatile frameworks in systems theory, enabling applications ranging from classical robust control to real-time safe operation of fleets of vehicles. While some optimal control problems can be efficiently solved using algebraic or convex methods, most general forms of optimal control must be solved using memory- expensive numerical methods. In this paper we present a theoretical formulation and a corresponding numerical algorithm that can find Pontryagin- optimal inputs for general dynamical systems by using a direct method. Pontryagin-optimal inputs, those satisfying the Minimum Principle, can be found for many classes of problems using indirect methods. But convergent numerical methods to solve indirect problems are hard to find and often converge slowly. On the other hand, convergent direct optimal control methods are fast and founded on solid theory, but their limit points are usually Banach-optimal inputs, which are a weaker form of optimality condition. Our result, founded on the theory of relaxed inputs as defined by J. Warga, establishes an equivalence between Pontryagin- optimal inputs and optimal relaxed inputs. Then, we formu- late a sampling-based numerical method to approximate the Pontryagin-optimal relaxed inputs using an iterative method. Finally, using a provably-convergent numerical method, we synthesize approximations of the Pontryagin-optimal inputs from the sampled relaxed inputs.

math.OC↗

arXiv2016

Gradient-Based Estimation of Air Flow and Geometry Configurations in a Building Using Fluid Dynamic Adjoint Equations

Real-time estimations of temperature distributions and geometric configurations are important to energy efficient buildings and the development of smarter cities. In this paper we formulate a gradient-based estimation algorithm capable of reconstructing the states of doors in a building, as well as its temperature distribution, based on a floor plan and a set of thermostats. Our algorithm solves in real time a convection-diffusion Computer Fluid Dynamics (CFD) model for the air flow in the building as a function of its geometric configuration. We formulate the estimation algorithm as an optimization problem, and we solve it by computing the adjoint equations of our CFD model, which we then use to obtain the gradients of the cost function with respect to the flow's temperature and door states. We evaluate the performance of our method using simulations of a real apartment in the St. Louis area. Our results show that the estimation method is both efficient and accurate, establishing its potential for the design of smarter control schemes in the operation of high-performance buildings.

math.OC↗

arXiv2015

Zoned HVAC Control via PDE-Constrained Optimization

Efficiency, comfort, and convenience are three major aspects in the design of control systems for residential Heating, Ventilation, and Air Conditioning (HVAC) units. In this paper we propose an optimization-based algorithm for HVAC control that minimizes energy consumption while maintaining a desired temperature in a room. Our algorithm uses a Computer Fluid Dynamics model, mathematically formulated using Partial Differential Equations (PDEs), to describe the interactions between temperature, pressure, and air flow. Our model allows us to naturally formulate problems such as controlling the temperature of a small region of interest within a room, or to control the speed of the air flow at the vents, which are hard to describe using finite-dimensional Ordinary Partial Differential (ODE) models. Our results show that our algorithm produces significant energy savings without a decrease in comfort.

math.OC↗