Dynamical Latent State Computation in the Posterior Parietal Cortex

Success in many real-world tasks depends on our ability to dynamically track hidden states of the world. To understand the underlying neural computations, we recorded brain activity in posterior parietal cortex (PPC) of monkeys navigating by optic flow to a hidden target location within a virtual environment, without explicit position cues. In addition to sequential neural dynamics and strong interneuronal interactions, we found that the hidden state -- monkey's displacement from the goal -- was encoded in single neurons, and could be dynamically decoded from population activity. The decoded estimates predicted navigation performance on individual trials. Task manipulations that perturbed the world model induced substantial changes in neural interactions, and modified the neural representation of the hidden state, while representations of sensory and motor variables remained stable. The findings were recapitulated by a task-optimized recurrent neural network model, suggesting that neural interactions in PPC embody the world model to consolidate information and track task-relevant hidden states.

The dynamics of cryptocurrency market behavior: sentiment analysis using Markov chains

PurposeThe authors examine cryptocurrency market behavior using a hidden Markov model (HMM). Under the assumption that the cryptocurrency market has unobserved heterogeneity, an HMM allows us to study (1) the extent to which cryptocurrency markets shift due to interactions with social sentiment during a bull or bear market and (2) the heterogeneous pattern of cryptocurrency market behavior under these two market conditions.Design/methodology/approachThe authors advance the HMM model based on two six-month datasets (from November 2017 to April 2018 for a bull market and from December 2018 to May 2019 for a bear market) collected from Google, Twitter, the stock market and cryptocurrency trading platforms in South Korea. Social sentiment data were collected by crawling Bitcoin-related posts on Twitter.FindingsThe authors highlight the reaction of the cryptocurrency market to social sentiment under a bull and a bear market and in two hidden states (an upward and a downward trend). They find: (1) social sentiment is relatively relevant during a bull compared to a bear market. (2) The cryptocurrency market in a downward state, that is, with a local decreasing trend, tends to be more responsive to positive social sentiment. (3) The market in an upward state, that is, with a local increasing trend, tends to better interact with negative social sentiment.Originality/valueThe proposed HMM model contributes to a theoretically grounded understanding of how cryptocurrency markets respond to social sentiment in bull and bear markets through varied sequences adjusted for cryptocurrency market heterogeneity.

Sequence-to-Sequence Acoustic Modeling with Semi-Stepwise Monotonic Attention for Speech Synthesis

An encoder–decoder with attention has become a popular method to achieve sequence-to-sequence (Seq2Seq) acoustic modeling for speech synthesis. To improve the robustness of the attention mechanism, methods utilizing the monotonic alignment between phone sequences and acoustic feature sequences have been proposed, such as stepwise monotonic attention (SMA). However, the phone sequences derived by grapheme-to-phoneme (G2P) conversion may not contain the pauses at the phrase boundaries in utterances, which challenges the assumption of strictly stepwise alignment in SMA. Therefore, this paper proposes to insert hidden states into phone sequences to deal with the situation that pauses are not provided explicitly, and designs a semi-stepwise monotonic attention (SSMA) to model these inserted hidden states. In this method, hidden states are introduced that absorb the pause segments in utterances in an unsupervised way. Thus, the attention at each decoding frame has three options, moving forward to the next phone, staying at the same phone, or jumping to a hidden state. Experimental results show that SSMA can achieve better naturalness of synthetic speech than SMA when phrase boundaries are not available. Moreover, the pause positions derived from the alignment paths of SSMA matched the manually labeled phrase boundaries quite well.

Hidden-state-only speciation and extinction models provide accurate tip estimates of diversification rates

1. State-dependent speciation and extinction (SSE) models provide a framework for testing potential correlations between the evolution of an observed trait and speciation and extinction rates. Recent expansions of these models allow for the inclusion of "hidden states" that, among other things, allow for rate heterogeneity often observed among lineages sharing a particular character state. However, in reality, multiple circumstances and interacting traits related to a focal character play a role in changing diversification dynamics of a lineage over time, restricting the use of available SSE models that require trait information to be assigned at the tips. 2. Here we introduce MiSSE, an SSE approach that infers diversification rate differences from hidden states only. It can be used similarly to other trait-free methods to estimate varying speciation, extinction, but also different functions of these parameters such as net-diversification, turnover rates, and extinction fraction. Given the size of the model space, we also describe an algorithm designed for efficiently searching through a reasonably large set of models without having to be exhaustive. 3. We compare the accuracy of rates inferred at the tips of the tree by MiSSE against popular character-free methods and demonstrate that the error associated with tip estimates is generally low. Due to certain characteristics of the SSE models, this method avoids some of the recent concerns with parameter identifiability in diversification analyses and can be used alongside regular phylogenetic comparative methods in trait-related diversification hypotheses. 4. Finally, we apply MiSSE, with a renewed focus on classic comparative methods, to understand processes happening near the present, rather than deep in the past, to examine how variation in plant height has impacted turnover rates in eucalypts, a species-rich lineage of flowering plants.

Queue hurdle Coxian phase-type model for two-stage process of population-based cancer screening

The quality assurance of two-stage population-based cancer screening program is determined by arrival rate (attending screening), positive rate (determined by the criteria of screening test), the compliance and the waiting time (WT) for confirmatory diagnosis in those screened as positive. These parameters were correlated between the process of screening procedures and the effectiveness of screening program. To capture such an inter-dependence of these parameters and quantify the effectiveness of program, we proposed a Queue hurdle Coxian phase-type (QH-CPH) model to estimate the arrival rate of screenees with the Poisson Queue process and the compliance rate of confirmatory diagnosis with the hurdle model, and also to identify the hidden states of WT that is affected by the capacity of health care and relevant covariates (such as demographic features and geographic areas) with the Coxian phase-type (CPH) process. We applied the proposed QH-CPH model to Taiwanese nationwide colorectal cancer screening program data for estimating the arrival rate and the probability of not complying with colonoscopy and classifying the compliers into two hidden states, short-waiting phase and long-waiting phase for colonoscopy. Significant covariates responsible for three processes were also identified by using the proportional hazards regression forms. A simulation study was further performed to assess the joint effect of these parameters on WT through a series of scenarios. The proposed QH-CPH model can provide an insight into the optimal and the practical design on population-based cancer screening for health policy-makers given the limited health care resources and capacity.

Vehicle Lane-Changes Trajectory Prediction Model Considering External Parameters

The ability to predict the motion of vehicles is essential for autonomous vehicles. Aiming at the problem that existing models cannot make full use of the external parameters including the outline of vehicles and the lane, we proposed a model to use the external parameters thoroughly when predicting the trajectory in the straight-line and non-free flow state. Meanwhile, dynamic sensitive area is proposed to filter out inconsequential surrounding vehicles. The historical trajectory of the vehicles and their external parameters are used as inputs. A shared Long Short-Term Memory (LSTM) cell is proposed to encode the explicit states obtained by mapping historical trajectory and external parameters. The hidden states of vehicles obtained from the last step are used to extract latent driving intent. Then, a convolution layer is designed to fuse hidden states to feed into the next prediction circle and a decoder is used to decode the hidden states of the vehicles to predict trajectory. The experiment result shows that the dynamic sensitive area can shorten the training time to 75.86% of the state-of-the-art work. Compared with other models, the accuracy of our model is improved by 23.7%. Meanwhile, the model's ability of anti-interference of external parameters is also improved.

Convolutional Recurrent Neural Networks for Text Classification

Recurrent neural network (RNN) and convolutional neural network (CNN) are two prevailing architectures used in text classification. Traditional approaches combine the strengths of these two networks by straightly streamlining them or linking features extracted from them. In this article, a novel approach is proposed to maintain the strengths of RNN and CNN to a great extent. In the proposed approach, a bi-directional RNN encodes each word into forward and backward hidden states. Then, a neural tensor layer is used to fuse bi-directional hidden states to get word representations. Meanwhile, a convolutional neural network is utilized to learn the importance of each word for text classification. Empirical experiments are conducted on several datasets for text classification. The superior performance of the proposed approach confirms its effectiveness.

Variational Beta Process Hidden Markov Models with Shared Hidden States for Trajectory Recognition

Hidden Markov model (HMM) is a vital model for trajectory recognition. As the number of hidden states in HMM is important and hard to be determined, many nonparametric methods like hierarchical Dirichlet process HMMs and Beta process HMMs (BP-HMMs) have been proposed to determine it automatically. Among these methods, the sampled BP-HMM models the shared information among different classes, which has been proved to be effective in several trajectory recognition scenes. However, the existing BP-HMM maintains a state transition probability matrix for each trajectory, which is inconvenient for classification. Furthermore, the approximate inference of the BP-HMM is based on sampling methods, which usually takes a long time to converge. To develop an efficient nonparametric sequential model that can capture cross-class shared information for trajectory recognition, we propose a novel variational BP-HMM model, in which the hidden states can be shared among different classes and each class chooses its own hidden states and maintains a unified transition probability matrix. In addition, we derive a variational inference method for the proposed model, which is more efficient than sampling-based methods. Experimental results on a synthetic dataset and two real-world datasets show that compared with the sampled BP-HMM and other related models, the variational BP-HMM has better performance in trajectory recognition.

Visual analytics tool for the interpretation of hidden states in recurrent neural networks

In this paper, we introduce a visual analytics approach aimed at helping machine learning experts analyze the hidden states of layers in recurrent neural networks. Our technique allows the user to interactively inspect how hidden states store and process information throughout the feeding of an input sequence into the network. The technique can help answer questions, such as which parts of the input data have a higher impact on the prediction and how the model correlates each hidden state configuration with a certain output. Our visual analytics approach comprises several components: First, our input visualization shows the input sequence and how it relates to the output (using color coding). In addition, hidden states are visualized through a nonlinear projection into a 2-D visualization space using t-distributed stochastic neighbor embedding to understand the shape of the space of the hidden states. Trajectories are also employed to show the details of the evolution of the hidden state configurations. Finally, a time-multi-class heatmap matrix visualizes the evolution of the expected predictions for multi-class classifiers, and a histogram indicates the distances between the hidden states within the original space. The different visualizations are shown simultaneously in multiple views and support brushing-and-linking to facilitate the analysis of the classifications and debugging for misclassified input sequences. To demonstrate the capability of our approach, we discuss two typical use cases for long short-term memory models applied to two widely used natural language processing datasets.