A5004775934- Neural dynamics and brain function
- Neural Networks and Applications
- Advanced Memory and Neural Computing
- Reinforcement Learning in Robotics
- Advanced Image Processing Techniques
- Functional Brain Connectivity Studies
- Image and Video Quality Assessment
- Artificial Intelligence in Games
- Advanced Graph Neural Networks
- Natural Language Processing Techniques
- Topic Modeling
- Image and Signal Denoising Methods
- Advanced Image and Video Retrieval Techniques
- Advanced Vision and Imaging
- Multimodal Machine Learning Applications
- Memory and Neural Mechanisms
- Mobile Crowdsensing and Crowdsourcing
- Fuel Cells and Related Materials
- Neurotransmitter Receptor Influence on Behavior
- Image Retrieval and Classification Techniques
- Speech and dialogue systems
- Neuroscience and Neuropharmacology Research
- Domain Adaptation and Few-Shot Learning
- Video Coding and Compression Technologies
- Software Engineering Research
Beijing University of Posts and Telecommunications
2016-2025
MRC Brain Network Dynamics Unit
2021-2024
University of Oxford
2018-2024
Medical Research Council
2022-2024
University of Liverpool
2024
Henan Agricultural University
2022
China University of Geosciences (Beijing)
2021
Mininglamp (China)
2020
Xinjiang University
2020
Changchun University of Science and Technology
2019
Panoramic video provides immersive and interactive experience by enabling humans to control the field of view (FoV) through head movement (HM). Thus, HM plays a key role in modeling human attention on panoramic video. This paper establishes database collecting subjects' sequences. From this database, we find that data are highly consistent across subjects. Furthermore, deep reinforcement learning (DRL) can be applied predict positions, via maximizing reward imitating scanpaths agent's...
Abstract For both humans and machines, the essence of learning is to pinpoint which components in its information processing pipeline are responsible for an error output, a challenge that known as ‘credit assignment’. It has long been assumed credit assignment best solved by backpropagation, also foundation modern machine learning. Here, we set out fundamentally different principle on called ‘prospective configuration’. In prospective configuration, network first infers pattern neural...
In this paper, a hybrid deep neural network scheduler (HDNNS) is proposed to solve job-shop scheduling problems (JSSPs). order mine the state information of schedule processing, problem divided into several classification-based subproblems. And learning framework used for solving these HDNNS applies convolution two-dimensional transformation method (CTDT) transform irregular regular features so that operation can be introduced dealing with JSSP. The simulation experiments designed testing...
The computational principles adopted by the hippocampus in associative memory (AM) tasks have been one of most studied topics and theoretical neuroscience. Recent theories suggested that AM predictive activities could be described within a unitary account, coding underlies computations supporting hippocampus. Following this theory, model based on classical hierarchical networks was proposed shown to perform well various tasks. However, fully did not incorporate recurrent connections, an...
The omnidirectional images (ODIs) are usually at low-resolution, due to the constraints of collection, storage and transmission. traditional two-dimensional (2D) image super-resolution methods not effective for spherical ODIs, because ODIs tend have non-uniformly distributed pixel density varying texture complexity across latitudes. In this work, we propose a novel latitude adaptive upscaling network (LAU-Net) ODI super-resolution, which allows pixels different latitudes adopt distinct...
Animals can adapt their preferences for different types of reward according to physiological state, such as hunger or thirst. To explain this ability, we employ a simple multi-objective reinforcement learning model that learns multiple values dimensions food water. We show by weighting these learned the current needs, behaviour may be flexibly adapted present preferences. This predicts individual dopamine neurons should encode errors associated with some more than others. provide preliminary...
Associative memories in the brain receive and store patterns of activity registered by sensory neurons, are able to retrieve them when necessary. Due their importance human intelligence, computational models associative have been developed for several decades now. In this paper, we present a novel neural model realizing memories, which is based on hierarchical generative network that receives external stimuli via neurons. It trained using predictive coding, an error-based learning algorithm...
Learning agents that are not only capable of taking tests, but also innovating is becoming a hot topic in AI. One the most promising paths towards this vision multi-agent learning, where act as environment for each other, and improving agent means proposing new problems others. However, existing evaluation platforms either compatible with settings, or limited to specific game. That is, there yet general platform research on intelligence. To end, we introduce Arena, intelligence 35 games...
The backpropagation of error algorithm (BP) used to train deep neural networks has been fundamental the successes learning. However, it requires sequential backwards updates and non-local computations which make challenging parallelize at scale is unlike how learning works in brain. Neuroscience-inspired algorithms, however, such as \emph{predictive coding} utilize local have potential overcome these limitations advance beyond technologies future. While predictive coding originated...
Deep learning has redefined AI thanks to the rise of artificial neural networks, which are inspired by neuronal networks in brain. Through years, these interactions between and neuroscience have brought immense benefits both fields, allowing be used a plethora applications. Neural use an efficient implementation reverse differentiation, called backpropagation (BP). This algorithm, however, is often criticized for its biological implausibility (e.g., lack local update rules parameters)....
Abstract Animals can adapt their preferences for different types reward according to physiological state, such as hunger or thirst. To describe this ability, we propose a simple extension of temporal difference model that learns multiple values each state dimensions food water. By weighting these learned the current needs, behaviour may be flexibly adapted present demands. Our predicts dopamine neurons should selective dimensions. We reanalysed data from primate and observed in addition...
Hierarchical reinforcement learning (HRL) has recently shown promising advances on speeding up learning, improving the exploration, and discovering intertask transferable skills. Most recent works focus HRL with two levels, i.e., a master policy manipulates subpolicies, which in turn manipulate primitive actions. However, multiple levels is usually needed many real-world scenarios, whose ultimate goals are highly abstract, while their actions very primitive. Therefore, this paper, we propose...
Predictive coding networks (PCNs) are an influential model for information processing in the brain. They have appealing theoretical interpretations and offer a single mechanism that accounts diverse perceptual phenomena of On other hand, backpropagation (BP) is commonly regarded to be most successful learning method modern machine learning. Thus, it exciting recent work formulates inference (IL) trains PCNs approximate BP. However, there several remaining critical issues: (i) IL...
Intrinsic rewards were introduced to simulate how human intelligence works; they are usually evaluated by intrinsically-motivated play, i.e., playing games without extrinsic but with rewards. However, none of the existing intrinsic reward approaches can achieve human-level performance under this very challenging setting play. In work, we propose a novel megalomania-driven (called mega-reward), which, our knowledge, is first approach that achieves in Intuitively, mega-reward comes from...
The study quantitatively described the effects of cinnamaldehyde on germination and growth Bacillus cereus spores in boiled ready-to-eat ground beef. With combination concentrations 0, 0.1, 0.5, 1.0% vol/wt at temperatures 12, 20, 28, 36°C, Huang model was successfully used as primary to predict lag time (λ) maximum rate (µmax). Thereafter, cubic polynomial models were estimate values Ln λ µmax considering both storage temperature concentration. highly accurate, because they produced...
The great success of deep learning has boosted the fast development video quality enhancement. However, existing methods mainly focus on enhancing objective compressed video, and ignore their perceptual that plays a key role in determining experience (QoE) videos. In this paper, we aim at video. Our main observation is enhancement mostly relies recovering high-frequency details with fine textures. Accordingly, propose novel generative adversarial network (GAN) based multi-level wavelet...
The automatic subpolicy discovery approach in hierarchical reinforcement learning (HRL) has recently achieved promising performance on sparse reward tasks. This accelerates transfer and unsupervised intelligent creatures while eliminating the domain-specific knowledge constraint. Most previously developed approaches are demonstrated to suffer from collapsing into situation where one dominates whole task, since they cannot ensure diversity of different subpolicies. In contrast, this article...
How the brain performs credit assignment is a fundamental unsolved problem in neuroscience. Many `biologically plausible' algorithms have been proposed, which compute gradients that approximate those computed by backpropagation (BP), and operate ways more closely satisfy constraints imposed neural circuitry. such utilize framework of energy-based models (EBMs), all free variables model are optimized to minimize global energy function. However, literature, these exist isolation no unified...
Abstract For both humans and machines, the essence of learning is to pinpoint which components in its information processing pipeline are responsible for an error output — a challenge that known as credit assignment . How brain solves key question neuroscience, also significant importance artificial intelligence. It has long been assumed best solved by backpropagation, foundation modern machine learning. However, it questioned whether possible implement backpropagation may actually be more...
Predictive coding (PC) is an influential theory in computational neuroscience, which argues that the cortex forms unsupervised world models by implementing a hierarchical process of prediction error minimization. PC networks (PCNs) are trained two phases. First, neural activities updated to optimize network's response external stimuli. Second, synaptic weights consolidate this change activity -- algorithm called \emph{prospective configuration}. While previous work has shown how various...