A5101904983- Machine Learning and Algorithms
- Gaussian Processes and Bayesian Inference
- Machine Learning and Data Classification
- Bayesian Methods and Mixture Models
- Thermodynamic properties of mixtures
- Reinforcement Learning in Robotics
- Phase Equilibria and Thermodynamics
- Domain Adaptation and Few-Shot Learning
- Neural Networks and Applications
- Markov Chains and Monte Carlo Methods
- Advanced Bandit Algorithms Research
- Chemical Thermodynamics and Molecular Structure
- Metaheuristic Optimization Algorithms Research
- Smart Grid Energy Management
- Robotics and Sensor-Based Localization
- Bayesian Modeling and Causal Inference
- Speech Recognition and Synthesis
- Microgrid Control and Optimization
- Data Stream Mining Techniques
- Stock Market Forecasting Methods
- Natural Language Processing Techniques
- Target Tracking and Data Fusion in Sensor Networks
- Advanced Multi-Objective Optimization Algorithms
- Electric Vehicles and Infrastructure
- Gas Sensing Nanomaterials and Sensors
Huaiyin Normal University
2025
Qingdao University of Science and Technology
2023-2024
Wuhan University
2024
Huzhou University
2024
Google (United Kingdom)
2016-2023
DeepMind (United Kingdom)
2016-2023
Xinjiang University
2022
Xi'an Jiaotong University
2016-2021
Google (United States)
2016-2021
PLA Army Engineering University
2017-2019
We learn recurrent neural network optimizers trained on simple synthetic functions by gradient descent. show that these learned exhibit a remarkable degree of transfer in they can be used to efficiently optimize broad range derivative-free black-box functions, including Gaussian process bandits, control objectives, global optimization benchmarks and hyper-parameter tuning tasks. Up the training horizon, trade-off exploration exploitation, compare favourably with heavily engineered Bayesian...
We present a meta-learning approach for adaptive text-to-speech (TTS) with few data. During training, we learn multi-speaker model using shared conditional WaveNet core and independent learned embeddings each speaker. The aim of training is not to produce neural network fixed weights, which then deployed as TTS system. Instead, the that requires data at deployment time rapidly adapt new speakers. introduce benchmark three strategies: (i) learning speaker embedding while keeping fixed, (ii)...
State-space models have been successfully used for more than fifty years in different areas of science and engineering. We present a procedure efficient variational Bayesian learning nonlinear state-space based on sparse Gaussian processes. The result is tractable posterior over dynamical systems. In comparison to conventional parametric models, we offer the possibility straightforwardly trade off model capacity computational cost whilst avoiding overfitting. Our main algorithm uses hybrid...
During the development of AlphaGo, its many hyper-parameters were tuned with Bayesian optimization multiple times. This automatic tuning process resulted in substantial improvements playing strength. For example, prior to match Lee Sedol, we latest AlphaGo agent and this improved win-rate from 50% 66.5% self-play games. version was deployed final match. Of course, since times during cycle, compounded contribution even higher than percentage. It is our hope that brief case study will be...
Genetic algorithms constitute a family of black-box optimization algorithms, which take inspiration from the principles biological evolution. While they provide general-purpose tool for optimization, their particular instantiations can be heuristic and motivated by loose intuition. In this work we explore fundamentally different approach: Given sufficiently flexible parametrization genetic operators, discover entirely new in data-driven fashion. More specifically, parametrize selection...
Optimizing functions without access to gradients is the remit of black-box methods such as evolution strategies. While highly general, their learning dynamics are often times heuristic and inflexible --- exactly limitations that meta-learning can address. Hence, we propose discover effective update rules for strategies via meta-learning. Concretely, our approach employs a search strategy parametrized by self-attention-based architecture, which guarantees rule invariant ordering candidate...
We introduce inference metaprogramming for probabilistic programming languages, including new language constructs, a formalism, and the rst demonstration of e ectiveness in practice. Instead relying on rigid black-box algorithms hard-coded into implementation as previous infer- ence enables developers to 1) dynamically decompose problems subproblems, 2) apply in- ference tactics 3) alternate between incorpo- rating data performing over existing data, 4) explore multiple execution traces...
Deep learning classifier selection has become a pivotal approach in hyperspectral image (HSI) classification, addressing the inherent challenges posed by high-dimensional data and limited labeled samples. Traditional methods often struggle with dynamic adaptability computational efficiency, especially under diverse complex distributions. To overcome these limitations, this study introduces two novel methods, Generative-Based Dynamic Ensemble of Heterogeneous Learning Classifiers (GDE-HDLC)...
Inertial odometry (IO) using only Measurement Units (IMUs) offers a lightweight and cost-effective solution for Unmanned Aerial Vehicle (UAV) applications, yet existing learning-based IO models often fail to generalize UAVs due the highly dynamic non-linear-flight patterns that differ from pedestrian motion. In this work, we identify conventional practice of transforming raw IMU data global coordinates undermines observability critical kinematic information in UAVs. By preserving body-frame...
Inertial odometry (IO) using only Measurement Units (IMUs) offers a lightweight and cost-effective solution for Unmanned Aerial Vehicle (UAV) applications, yet existing learning-based IO models often fail to generalize UAVs due the highly dynamic non-linear-flight patterns that differ from pedestrian motion. In this work, we identify conventional practice of transforming raw IMU data global coordinates undermines observability critical kinematic information in UAVs. By preserving body-frame...
We extend the herding algorithm to continuous spaces by using kernel trick. The resulting "kernel herding" is an infinite memory deterministic process that learns approximate a PDF with collection of samples. show decreases error expectations functions in Hilbert space at rate O(1/T) which much faster than usual O(1/pT) for iid random illustrate approximating Bayesian predictive distributions.
Off-policy evaluation (OPE) holds the promise of being able to leverage large, offline datasets for both evaluating and selecting complex policies decision making. The ability learn is particularly important in many real-world domains, such as healthcare, recommender systems, or robotics, where online data collection an expensive potentially dangerous process. Being accurately evaluate select high-performing without requiring interaction could yield significant benefits safety, time, cost...
A simple thermal-conversion strategy enabled high-efficiency MgO/NPC microwave absorbers with different morphologies and distinct conductivity to be formed.