Article
Computer Science, Artificial Intelligence
Hongkui Wang, Li Yu, Junhui Liang, Haibing Yin, Tiansong Li, Shengwei Wang
Summary: This paper models the human visual system (HVS) as a three-level communication system based on the hierarchical predictive coding theory, proposes a novel just noticeable distortion (JND) estimation scheme, and describes the effects of neural responses and surrounding residues on positive and negative perception. The proposed method measures perception effects based on surprise, with the JND threshold of each stage estimated individually and the total JND threshold obtained through non-linear superposition. Incorporating the JND estimation scheme into Versatile Video Coding codec for image compression shows significant bit rate reduction without compromising perceptual quality.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Theory & Methods
Junyi Zhang, Angelos Dassios
Summary: This paper presents a new random probability measure called the truncated Poisson-Dirichlet process. It introduces a finite approximation for the distribution of the Dirichlet process by truncating the components in descending order according to their random weights. The proposed method has a lower truncation error compared to existing stick-breaking processes.
STATISTICS AND COMPUTING
(2023)
Article
Psychology, Multidisciplinary
Magdalena Lhotka, Anja Ischebeck, Birgit Helmlinger, Natalia Zaretskaya
Summary: Predictive coding theory proposes that subjective experience is the result of comparing sensory input with top-down predictions. This theory explains various phenomena, including visual illusions, hallucinations, and psychosis. A study tested the connection between these phenomena and found a positive association between pareidolia proneness and delusional ideations, supporting the hierarchical view of predictive processing.
FRONTIERS IN PSYCHOLOGY
(2023)
Article
Automation & Control Systems
Marta Catalano, Pierpaolo De Blasi, Antonio Lijoi, Igor Prunster
Summary: Bayesian hierarchical models are powerful tools for learning common latent features across multiple data sources. This study establishes theoretical guarantees for recovering the true data generating process in the Hierarchical Dirichlet Process (HDP) or a generalization of the HDP. The posterior contraction rates are affected by the relationship between sample sizes.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Computer Science, Information Systems
Weijian Ni, Gang Zhao, Tong Liu, Qingtian Zeng, Xingzong Xu
Summary: Predictive business process monitoring predicts the next stage of a business process based on past events, leading to better resource allocation and improved execution efficiency. A hierarchical Transformer-based model is proposed to enhance the performance of predictive monitoring. It uses different encoding methods to capture the relationship between activities and attributes, incorporates a drift detection algorithm, utilizes cross-attention to calculate correlations, and incorporates learnable position encoding for subsequence information. Experimental results show an average improvement of 6.32% in next activity prediction accuracy and a 21% reduction in mean absolute error for remaining time prediction.
Article
Computer Science, Artificial Intelligence
Abdulrahman Jalayer, Mohsen Kahani, Asef Pourmasoumi, Amin Beheshti
Summary: Predictive business process monitoring is an important task in business process management, but traditional neural networks may not be suitable due to their limitations in considering event order and attribute importance. This study proposes a new approach based on LSTM and attention mechanism, which performs well in predicting the next activity of an ongoing process according to experimental evaluation of real-world event logs.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Interdisciplinary Applications
Diogo Ferrari
Summary: The existence of latent clusters with different responses to treatment is a significant concern in scientific research. This article discusses the implementation of a novel hierarchical Dirichlet process approach using the R package hdpGLM. The methods provided in the package make it easier for researchers to investigate heterogeneity in treatment effects and identify clusters of subjects with differential effects.
JOURNAL OF STATISTICAL SOFTWARE
(2023)
Article
Mathematical & Computational Biology
William H. Alexander, Thilo Womelsdorf
Summary: The interaction between medial and lateral prefrontal cortex plays a crucial role in cognitive control and decision-making, with proposals suggesting complementary roles in different aspects of behavior. The Hierarchical Error Representation model places these regions within the framework of predictive coding, demonstrating how they interact during behavioral periods. This model is able to capture neurophysiological, behavioral, and network effects, providing evidence for predictive coding as a unifying framework for understanding PFC function.
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2021)
Article
Ethics
Alessio Tacca, Frederic Gilbert
Summary: This article discusses the applications of predictive neurotechnologies in advisory devices, the ethical concerns of relying on predictive neural devices, and the risks of over-dependence on technology for users. The concept of epistemic authority is explored, and the relationship between predictive devices and users is examined.
Review
Neurosciences
Zhe Sage Chen
Summary: Predictive coding is a computational theory used to describe how the brain perceives and acts. It has been widely used in sensory processing and motor control. In the context of nociceptive and pain processing, the cingulate cortex and insula cortex are two major hubs that mediate information from sensory afferents and spinothalamic inputs. This mini-review presents an updated hierarchical predictive coding framework for pain perception and discusses its computational, algorithmic, and implementation issues, proposing active inference as a generalized predictive coding algorithm and hierarchically organized traveling waves of independent neural oscillations as a plausible brain mechanism for integrating bottom-up and top-down information across distributed pain circuits.
FRONTIERS IN NEURAL CIRCUITS
(2023)
Article
Biology
Gregory Faye, Guilhem Fouilhe, Rufin VanRullen
Summary: Sensory perception relies on hierarchical cortical areas and neural activity propagates bidirectionally to convey information about sensory inputs, cognitive states, expectations, and predictions. This study presents a mathematical framework to investigate neural dynamics in a hierarchical perceptual system and shows that stability and propagation speed can be determined by controlling the hyperparameters of different signals. Different neural assemblies can display distinct properties in terms of stability, propagation speed, and direction. The study also analyzes the influence of transmission delays and reveals the emergence of oscillations.
BULLETIN OF MATHEMATICAL BIOLOGY
(2023)
Article
Automation & Control Systems
Hong-Gui Han, Shi-Jia Fu, Hao-Yuan Sun, Jun-Fei Qiao
Summary: A hierarchical nonlinear model predictive control (HNMPC) strategy is developed to address the different time scales of controlled variables in wastewater treatment process (WWTP) for improved operation performance. The main advantages of HNMPC include the development of a hierarchical control structure, a gradient method to solve the optimization problem, and theoretical proof of stability. Testing results on a benchmark simulation model confirm the proposed HNMPC's suitable operation performance in terms of control accuracy.
JOURNAL OF PROCESS CONTROL
(2021)
Article
Statistics & Probability
Shui Feng
Summary: The Hierarchical Dirichlet process is a crucial Bayesian non-parametric prior for studying clustered data groups. It models each group using a level two Dirichlet process, and all groups share the same base distribution generated from a level one Dirichlet process. This paper presents the main results of the law of large numbers and large deviations for the hierarchical Dirichlet process, as well as its mass, when both concentration parameters tend to infinity. The explicit identification of large deviation rate functions shows that the rate function for the hierarchical Dirichlet process is composed of two terms corresponding to the relative entropies at each level. It is lower than the rate function for the Dirichlet process, indicating a slower growth rate of the number of clusters under the hierarchical Dirichlet process.
ELECTRONIC COMMUNICATIONS IN PROBABILITY
(2023)
Article
Ethics
Julia Haas
Summary: This paper explores the limitations of hierarchical predictive coding in explaining binocular rivalry, and proposes a modified version of predictive processing to account for the role of reward in this phenomenon. However, accepting this explanation may contradict the epistemic commitments favored by proponents of hierarchical predictive coding.
PHILOSOPHICAL PSYCHOLOGY
(2021)
Article
Computer Science, Information Systems
Koffi Eddy Ihou, Manar Amayri, Nizar Bouguila
Summary: This article introduces a Bayesian nonparametric (BNP) approach to handle model selection and topic sharing in topic models. By applying the Hierarchical Dirichlet process (HDP) to the BNP topic model, the model is able to more effectively characterize dependencies between documents and generate more robust and realistic compression algorithms.
ACM TRANSACTIONS ON KNOWLEDGE DISCOVERY FROM DATA
(2022)