scholarly journals Efficient Hyperparameter Optimization for Physics-based Character Animation

2021 ◽  
Vol 4 (1) ◽  
pp. 1-19
Author(s):  
Zeshi Yang ◽  
Zhiqi Yin
Keyword(s):  
Control Systems ◽  
Task Difficulty ◽  
State Of The Art ◽  
Optimization Methods ◽  
Search Space ◽  
Character Animation ◽  
Bayesian Optimization ◽  
Efficiency Gain ◽  
Hyperparameter Optimization ◽  
Optimization Framework

Physics-based character animation has seen significant advances in recent years with the adoption of Deep Reinforcement Learning (DRL). However, DRL-based learning methods are usually computationally expensive and their performance crucially depends on the choice of hyperparameters. Tuning hyperparameters for these methods often requires repetitive training of control policies, which is even more computationally prohibitive. In this work, we propose a novel Curriculum-based Multi-Fidelity Bayesian Optimization framework (CMFBO) for efficient hyperparameter optimization of DRL-based character control systems. Using curriculum-based task difficulty as fidelity criterion, our method improves searching efficiency by gradually pruning search space through evaluation on easier motor skill tasks. We evaluate our method on two physics-based character control tasks: character morphology optimization and hyperparameter tuning of DeepMimic. Our algorithm significantly outperforms state-of-the-art hyperparameter optimization methods applicable for physics-based character animation. In particular, we show that hyperparameters optimized through our algorithm result in at least 5x efficiency gain comparing to author-released settings in DeepMimic.

scholarly journals Efficient Automatic CASH via Rising Bandits

2020 ◽  
Vol 34 (04) ◽  
pp. 4763-4771
Author(s):  
Yang Li ◽  
Jiawei Jiang ◽  
Jinyang Gao ◽  
Yingxia Shao ◽  
Ce Zhang ◽  
...  
Keyword(s):  
Machine Learning ◽  
Online Algorithm ◽  
Automatic Machine ◽  
Bayesian Optimization ◽  
Algorithm Selection ◽  
Hyperparameter Optimization ◽  
Alternating Optimization ◽  
Optimization Framework ◽  
Low Efficiency ◽  
Combined Algorithm

The Combined Algorithm Selection and Hyperparameter optimization (CASH) is one of the most fundamental problems in Automatic Machine Learning (AutoML). The existing Bayesian optimization (BO) based solutions turn the CASH problem into a Hyperparameter Optimization (HPO) problem by combining the hyperparameters of all machine learning (ML) algorithms, and use BO methods to solve it. As a result, these methods suffer from the low-efficiency problem due to the huge hyperparameter space in CASH. To alleviate this issue, we propose the alternating optimization framework, where the HPO problem for each ML algorithm and the algorithm selection problem are optimized alternately. In this framework, the BO methods are used to solve the HPO problem for each ML algorithm separately, incorporating a much smaller hyperparameter space for BO methods. Furthermore, we introduce Rising Bandits, a CASH-oriented Multi-Armed Bandits (MAB) variant, to model the algorithm selection in CASH. This framework can take the advantages of both BO in solving the HPO problem with a relatively small hyperparameter space and the MABs in accelerating the algorithm selection. Moreover, we further develop an efficient online algorithm to solve the Rising Bandits with provably theoretical guarantees. The extensive experiments on 30 OpenML datasets demonstrate the superiority of the proposed approach over the competitive baselines.

scholarly journals Binarized Neural Architecture Search

2020 ◽  
Vol 34 (07) ◽  
pp. 10526-10533 ◽  
Author(s):  
Hanlin Chen ◽  
Li'an Zhuo ◽  
Baochang Zhang ◽  
Xiawu Zheng ◽  
Jianzhuang Liu ◽  
...  
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
State Of The Art ◽  
Optimization Methods ◽  
Search Space ◽  
Network Architectures ◽  
Neural Architecture ◽  
Space Reduction ◽  
The Cost ◽  
A Performance

Neural architecture search (NAS) can have a significant impact in computer vision by automatically designing optimal neural network architectures for various tasks. A variant, binarized neural architecture search (BNAS), with a search space of binarized convolutions, can produce extremely compressed models. Unfortunately, this area remains largely unexplored. BNAS is more challenging than NAS due to the learning inefficiency caused by optimization requirements and the huge architecture space. To address these issues, we introduce channel sampling and operation space reduction into a differentiable NAS to significantly reduce the cost of searching. This is accomplished through a performance-based strategy used to abandon less potential operations. Two optimization methods for binarized neural networks are used to validate the effectiveness of our BNAS. Extensive experiments demonstrate that the proposed BNAS achieves a performance comparable to NAS on both CIFAR and ImageNet databases. An accuracy of 96.53% vs. 97.22% is achieved on the CIFAR-10 dataset, but with a significantly compressed model, and a 40% faster search than the state-of-the-art PC-DARTS.

Learning manipulation skills from a single demonstration

2017 ◽  
Vol 37 (1) ◽  
pp. 137-154 ◽  
Author(s):  
Peter Englert ◽  
Marc Toussaint
Keyword(s):  
State Of The Art ◽  
Optimization Methods ◽  
Black Box ◽  
Bayesian Optimization ◽  
High Dimensional ◽  
Control Methods ◽  
Benchmark Experiment ◽  
Real Robot ◽  
Optimal Control Methods ◽  
Single Demonstration

We consider the scenario where a robot is demonstrated a manipulation skill once and should then use only a few trials on its own to learn to reproduce, optimize, and generalize that same skill. A manipulation skill is generally a high-dimensional policy. To achieve the desired sample efficiency, we need to exploit the inherent structure in this problem. With our approach, we propose to decompose the problem into analytically known objectives, such as motion smoothness, and black-box objectives, such as trial success or reward, depending on the interaction with the environment. The decomposition allows us to leverage and combine (i) constrained optimization methods to address analytic objectives, (ii) constrained Bayesian optimization to explore black-box objectives, and (iii) inverse optimal control methods to eventually extract a generalizable skill representation. The algorithm is evaluated on a synthetic benchmark experiment and compared with state-of-the-art learning methods. We also demonstrate the performance on real-robot experiments with a PR2.

scholarly journals Weighted Random Search for CNN Hyperparameter Optimization

2020 ◽  
Vol 15 (2) ◽  
Author(s):  
Razvan Andonie ◽  
Adrian-Catalin Florea
Keyword(s):  
Neural Network ◽  
Classification Accuracy ◽  
State Of The Art ◽  
Random Search ◽  
Optimization Methods ◽  
Optimal Combination ◽  
The Other ◽  
Greedy Heuristic ◽  
Training Phase ◽  
Hyperparameter Optimization

Nearly all model algorithms used in machine learning use two different sets of parameters: the training parameters and the meta-parameters (hyperparameters). While the training parameters are learned during the training phase, the values of the hyperparameters have to be specified before learning starts. For a given dataset, we would like to find the optimal combination of hyperparameter values, in a reasonable amount of time. This is a challenging task because of its computational complexity. In previous work [11], we introduced the Weighted Random Search (WRS) method, a combination of Random Search (RS) and probabilistic greedy heuristic. In the current paper, we compare the WRS method with several state-of-the art hyperparameter optimization methods with respect to Convolutional Neural Network (CNN) hyperparameter optimization. The criterion is the classification accuracy achieved within the same number of tested combinations of hyperparameter values. According to our experiments, the WRS algorithm outperforms the other methods.

scholarly journals Exploiting Partial Assignments for Efficient Evaluation of Answer Set Programs with External Source Access

2018 ◽  
Vol 62 ◽  
pp. 665-727 ◽  
Author(s):  
Thomas Eiter ◽  
Tobias Kaminski ◽  
Christoph Redl ◽  
Antonius Weinzierl
Keyword(s):  
State Of The Art ◽  
Search Space ◽  
External Source ◽  
The State ◽  
Efficiency Gain ◽  
Model Search ◽  
Wide Range ◽  
Search Space Pruning ◽  
Answer Sets ◽  
Answer Set

Answer Set Programming (ASP) is a well-known declarative problem solving approach based on nonmonotonic logic programs, which has been successfully applied to a wide range of applications in artificial intelligence and beyond. To address the needs of modern applications, HEX-programs were introduced as an extension of ASP with external atoms for accessing information outside programs via an API style bi-directional interface mechanism. To evaluate such programs, conflict-driving learning algorithms for SAT and ASP solving have been extended in order to capture the semantics of external atoms. However, a drawback of the state-of-the-art approach is that external atoms are only evaluated under complete assignments (i.e., input to the external source) while in practice, their values often can be determined already based on partial assignments alone (i.e., from incomplete input to the external source). This prevents early backtracking in case of conflicts, and hinders more efficient evaluation of HEX-programs. We thus extend the notion of external atoms to allow for three-valued evaluation under partial assignments, while the two-valued semantics of the overall HEX-formalism remains unchanged. This paves the way for three enhancements: first, to evaluate external sources at any point during model search, which can trigger learning knowledge about the source behavior and/or early backtracking in the spirit of theory propagation in SAT modulo theories (SMT). Second, to optimize the knowledge learned in terms of so-called nogoods, which roughly speaking are impossible input-output configurations. Shrinking nogoods to their relevant input part leads to more effective search space pruning. And third, to make a necessary minimality check of candidate answer sets more efficient by exploiting early external evaluation calls. As this check usually accounts for a large share of the total runtime, optimization is here particularly important. We further present an experimental evaluation of an implementation of a novel HEX-algorithm that incorporates these enhancements using a benchmark suite. Our results demonstrate a clear efficiency gain over the state-of-the-art HEX-solver for the benchmarks, and provide insights regarding the most effective combinations of solver configurations.

scholarly journals Using MAP-Elites to Optimize Self-Assembling Behaviors in a Swarm of Bio-micro-robots

2019 ◽  
Author(s):  
Leo Cazenille ◽  
Nicolas Bredeche ◽  
Nathanael Aubert-Kato
Keyword(s):  
Self Assembly ◽  
State Of The Art ◽  
Optimization Methods ◽  
Search Space ◽  
Reaction Networks ◽  
High Performing ◽  
Robot Swarms ◽  
Self Assembling ◽  
Previous State ◽  
Molecular Robots

AbstractWe are interested in programming a swarm of molecular robots that can perform self-assembly to form specific shapes at a specific location. Programming such robot swarms is challenging for two reasons. First, the goal is to optimize both the parameters and the structure of chemical reaction networks. Second, the search space is both high-dimensional and deceptive. In this paper, we show that MAP-Elites, an algorithm that searches for both high-performing and diverse solutions, outperforms previous state-of-the-art optimization methods.

scholarly journals Hyperparameter Optimization Techniques for Designing Software Sensors Based on Artificial Neural Networks

Sensors ◽  
2021 ◽  
Vol 21 (24) ◽  
pp. 8435
Author(s):  
Sebastian Blume ◽  
Tim Benedens ◽  
Dieter Schramm
Keyword(s):  
Genetic Algorithm ◽  
Optimization Methods ◽  
Optimization Techniques ◽  
Bayesian Optimization ◽  
Training Process ◽  
Hyperparameter Optimization ◽  
Data Set ◽  
Knowledge Based ◽  
Software Sensors ◽  
Artificial Neural

Software sensors are playing an increasingly important role in current vehicle development. Such soft sensors can be based on both physical modeling and data-based modeling. Data-driven modeling is based on building a model purely on captured data which means that no system knowledge is required for the application. At the same time, hyperparameters have a particularly large influence on the quality of the model. These parameters influence the architecture and the training process of the machine learning algorithm. This paper deals with the comparison of different hyperparameter optimization methods for the design of a roll angle estimator based on an artificial neural network. The comparison is drawn based on a pre-generated simulation data set created with ISO standard driving maneuvers. Four different optimization methods are used for the comparison. Random Search and Hyperband are two similar methods based purely on randomness, whereas Bayesian Optimization and the genetic algorithm are knowledge-based methods, i.e., they process information from previous iterations. The objective function for all optimization methods consists of the root mean square error of the training process and the reference data generated in the simulation. To guarantee a meaningful result, k-fold cross-validation is integrated for the training process. Finally, all methods are applied to the predefined parameter space. It is shown that the knowledge-based methods lead to better results. In particular, the Genetic Algorithm leads to promising solutions in this application.

scholarly journals Using Graphs of Classifiers to Impose Declarative Constraints on Semi-supervised Learning

2017 ◽  
Author(s):  
Lidong Bing ◽  
William W. Cohen ◽  
Bhuwan Dhingra
Keyword(s):  
Supervised Learning ◽  
Text Classification ◽  
Supervised Classification ◽  
State Of The Art ◽  
Relation Extraction ◽  
Optimization Methods ◽  
Bayesian Optimization ◽  
Declarative Language ◽  
Domain Specific ◽  
Classification Tasks

We propose a general approach to modeling semi-supervised learning (SSL) algorithms. Specifically, we present a declarative language for modeling both traditional supervised classification tasks and many SSL heuristics, including both well-known heuristics such as co-training and novel domain-specific heuristics. In addition to representing individual SSL heuristics, we show that multiple heuristics can be automatically combined using Bayesian optimization methods. We experiment with two classes of tasks, link-based text classification and relation extraction. We show modest improvements on well-studied link-based classification benchmarks, and state-of-the-art results on relation-extraction tasks for two realistic domains.

scholarly journals Multi-fidelity black-box optimization for time-optimal quadrotor maneuvers

2021 ◽  
pp. 027836492110333
Author(s):  
Gilhyun Ryou ◽  
Ezra Tal ◽  
Sertac Karaman
Keyword(s):  
Real World ◽  
High Speed ◽  
Analytical Approximation ◽  
Black Box ◽  
Bayesian Optimization ◽  
Optimization Framework ◽  
Quadrotor Aircraft ◽  
Time Optimal ◽  
Feasibility Constraints ◽  
Generation Problem

We consider the problem of generating a time-optimal quadrotor trajectory for highly maneuverable vehicles, such as quadrotor aircraft. The problem is challenging because the optimal trajectory is located on the boundary of the set of dynamically feasible trajectories. This boundary is hard to model as it involves limitations of the entire system, including complex aerodynamic and electromechanical phenomena, in agile high-speed flight. In this work, we propose a multi-fidelity Bayesian optimization framework that models the feasibility constraints based on analytical approximation, numerical simulation, and real-world flight experiments. By combining evaluations at different fidelities, trajectory time is optimized while the number of costly flight experiments is kept to a minimum. The algorithm is thoroughly evaluated for the trajectory generation problem in two different scenarios: (1) connecting predetermined waypoints; (2) planning in obstacle-rich environments. For each scenario, we conduct both simulation and real-world flight experiments at speeds up to 11 m/s. Resulting trajectories were found to be significantly faster than those obtained through minimum-snap trajectory planning.

scholarly journals MLatom 2: An Integrative Platform for Atomistic Machine Learning

2021 ◽  
Vol 379 (4) ◽  
Author(s):  
Pavlo O. Dral ◽  
Fuchun Ge ◽  
Bao-Xin Xue ◽  
Yi-Fan Hou ◽  
Max Pinheiro ◽  
...  
Keyword(s):  
Machine Learning ◽  
Software Package ◽  
State Of The Art ◽  
Kernel Method ◽  
Automatic Learning ◽  
Hyperparameter Optimization ◽  
Input And Output ◽  
Spectrum Simulation ◽  
Farthest Point ◽  
Theoretical Foundations

AbstractAtomistic machine learning (AML) simulations are used in chemistry at an ever-increasing pace. A large number of AML models has been developed, but their implementations are scattered among different packages, each with its own conventions for input and output. Thus, here we give an overview of our MLatom 2 software package, which provides an integrative platform for a wide variety of AML simulations by implementing from scratch and interfacing existing software for a range of state-of-the-art models. These include kernel method-based model types such as KREG (native implementation), sGDML, and GAP-SOAP as well as neural-network-based model types such as ANI, DeepPot-SE, and PhysNet. The theoretical foundations behind these methods are overviewed too. The modular structure of MLatom allows for easy extension to more AML model types. MLatom 2 also has many other capabilities useful for AML simulations, such as the support of custom descriptors, farthest-point and structure-based sampling, hyperparameter optimization, model evaluation, and automatic learning curve generation. It can also be used for such multi-step tasks as Δ-learning, self-correction approaches, and absorption spectrum simulation within the machine-learning nuclear-ensemble approach. Several of these MLatom 2 capabilities are showcased in application examples.

Sign in / Sign up

Export Citation Format

Share Document