We present a novel recommendation system designed to provide real-time treatment strategies to therapists during psychotherapy sessions. Our system utilizes a turn-level rating mechanism that forecasts the therapeutic outcome by calculating a similarity score between the profound representation of a scoring inventory and the patient's current spoken sentence. By transcribing and segmenting the continuous audio stream into patient and therapist turns, our system conducts immediate evaluation of their therapeutic working alliance. The resulting dialogue pairs, along with their computed working alliance ratings, are then utilized in a deep reinforcement learning recommendation system. In this system, the sessions are treated as users, while the topics are treated as items. To showcase the system's effectiveness, we not only evaluate its performance using an existing dataset of psychotherapy sessions but also demonstrate its practicality through a web app. Through this demo, we aim to provide a tangible and engaging experience of our recommendation system in action.

Information-theoretic clustering is one of the most promising and principled approaches to finding clusters with minimal apriori assumptions. The key criterion therein is to maximize the mutual information between the data points and their cluster labels. Such an approach, however, does not explicitly promote any type of inter-cluster behavior. We instead propose to maximize the Kullback-Leibler divergence between the underlying data distributions associated to clusters (referred to as cluster distributions). We show it to entail the mutual information criterion along with maximizing cross entropy between the cluster distributions. For practical efficiency, we propose to empirically estimate the objective of KL-D between clusters in its dual form leveraging deep neural nets as a dual function approximator. Remarkably, our theoretical analysis establishes that estimating the divergence measure in its dual form simplifies the problem of clustering to one of optimally finding k-1 cut points for k clusters in the 1-D dual functional space. Overall, our approach enables linear-time clustering algorithms with theoretical guarantees of near-optimality, owing to the submodularity of the objective. We show the empirical superiority of our approach w.r.t. current state-of-the-art methods on the challenging task of clustering noisy timeseries as observed in domains such as neuroscience, healthcare, financial markets, spatio-temporal environmental dynamics, etc.

Unlike traditional time series, the action sequences of human decision making usually involve many cognitive processes such as beliefs, desires, intentions and theory of mind, i.e. what others are thinking. This makes predicting human decision making challenging to be treated agnostically to the underlying psychological mechanisms. We propose to use a recurrent neural network architecture based on long short-term memory networks (LSTM) to predict the time series of the actions taken by the human subjects at each step of their decision making, the first application of such methods in this research domain. In this study, we collate the human data from 8 published literature of the Iterated Prisoner's Dilemma comprising 168,386 individual decisions and postprocess them into 8,257 behavioral trajectories of 9 actions each for both players. Similarly, we collate 617 trajectories of 95 actions from 10 different published studies of Iowa Gambling Task experiments with healthy human subjects. We train our prediction networks on the behavioral data from these published psychological experiments of human decision making, and demonstrate a clear advantage over the state-of-the-art methods in predicting human decision making trajectories in both single-agent scenarios such as the Iowa Gambling Task and multi-agent scenarios such as the Iterated Prisoner's Dilemma. In the prediction, we observe that the weights of the top performers tends to have a wider distribution, and a bigger bias in the LSTM networks, which suggests possible interpretations for the distribution of strategies adopted by each group.11The data and codes to reproduce the empirical results can be accessed and reproduced at https://github.com/doerlbh/HumanLST M. A extended version of this work is to appear in [1].

Many natural systems, especially biological ones, exhibit complex multivariate nonlinear dynamical behaviors that can be hard to capture by linear autoregressive models. On the other hand, generic nonlinear models such as deep recurrent neural networks often require large amounts of training data, not always available in domains such as brain imaging; also, they often lack interpretability. Domain knowledge about the types of dynamics typically observed in such systems, such as a certain type of dynamical systems models, could complement purely data-driven techniques by providing a good prior. In this work, we consider a class of ordinary differential equation (ODE) models known as van der Pol (VDP) oscil lators and evaluate their ability to capture a low-dimensional representation of neural activity measured by different brain imaging modalities, such as calcium imaging (CaI) and fMRI, in different living organisms: larval zebrafish, rat, and human. We develop a novel and efficient approach to the nontrivial problem of parameters estimation for a network of coupled dynamical systems from multivariate data and demonstrate that the resulting VDP models are both accurate and interpretable, as VDP's coupling matrix reveals anatomically meaningful excitatory and inhibitory interactions across different brain subsystems. VDP outperforms linear autoregressive models (VAR) in terms of both the data fit accuracy and the quality of insight provided by the coupling matrices and often tends to generalize better to unseen data when predicting future brain activity, being comparable to and sometimes better than the recurrent neural networks (LSTMs). Finally, we demonstrate that our (generative) VDP model can also serve as a data-augmentation tool leading to marked improvements in predictive accuracy of recurrent neural networks. Thus, our work contributes to both basic and applied dimensions of neuroimaging: gaining scientific insights and improving brain-based predictive models, an area of potentially high practical importance in clinical diagnosis and neurotechnology.

We propose a novel dialogue modeling framework, the first-ever nonparametric kernel functions based approach for dialogue modeling, which learns hashcodes as text representations; unlike traditional deep learning models, it handles well relatively small datasets, while also scaling to large ones. We also derive a novel lower bound on mutual information, used as a model-selection criterion favoring representations with better alignment between the utterances of participants in a collaborative dialogue setting, as well as higher predictability of the generated responses. As demonstrated on three real-life datasets, including prominently psychotherapy sessions, the proposed approach significantly outperforms several state-of-art neural network based dialogue systems, both in terms of computational efficiency, reducing training time from days or weeks to hours, and the response quality, achieving an order of magnitude improvement over competitors in frequency of being chosen as the best model by human evaluators.

Kernel methods have produced state-of-the-art results for a number of NLP tasks such as relation extraction, but suffer from poor scalability due to the high cost of computing kernel similarities between natural language structures. A recently proposed technique, kernelized locality-sensitive hashing (KLSH), can significantly reduce the computational cost, but is only applicable to classifiers operating on kNN graphs. Here we propose to use random subspaces of KLSH codes for efficiently constructing an explicit representation of NLP structures suitable for general classification methods. Further, we propose an approach for optimizing the KLSH model for classification problems by maximizing an approximation of mutual information between the KLSH codes (feature vectors) and the class labels. We evaluate the proposed approach on biomedical relation extraction datasets, and observe significant and robust improvements in accuracy w.r.t. state-ofthe-art classifiers, along with drastic (orders-of-magnitude) speedup compared to conventional kernel methods.

Sahil Garg, Aram Galstyan, Greg Ver Steeg, Guillermo Cecchi. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

One of the main foci of addiction research is the delineation of markers that track the propensity of relapse. Speech analysis can provide an unbiased assessment that can be deployed outside the lab, enabling objective measurements and relapse susceptibility tracking. This work is the first attempt to study unscripted speech markers in cocaine users. We analyzed 23 subjects performing two tasks: describing the positive consequences (PC) of abstinence and the negative consequences (NC) of using cocaine. We perform two main experiments: first, we analyzed whether acoustic and semantic features can infer clinical variables such as the Cocaine Selective Severity Assessment; then, we analyzed the main problem of interest: to see if these features are powerful enough to infer if the subjects remains abstinent. Our results show that speech features have potential to be used as a proxy to monitor cocaine users under treatment to recover from their addiction.

How do people hear laughter in expressive, unprompted speech? What is the range of expressivity and function of laughter in this speech, and how can laughter inform the recognition of higher-level expressive dimensions in a corpus? This paper presents a scalable method for collecting natural human description of laughter, transforming the description to a vector of quantifiable laughter dimensions, and deriving baseline classifiers for the different dimensions of expressive laughter. Then, it explores the impact of leveraging nuances of laughter in the recognition of higher-level, general expressive dimensions, discovered in the same way, such as genuine happiness, sarcasm, nervous reflection, and more. The performance of the low-level laughter classifiers is presented, along with the performance of the high-level laughter-aware and laughter-unaware classifiers.

Prototypical examples that best summarize and compactly represent an underlying complex data distribution, communicate meaningful insights to humans in domains where simple explanations are hard to extract. In this paper, we present algorithms with strong theoretical guarantees to mine these data sets and select prototypes, a.k.a. representatives that optimally describes them. Our work notably generalizes the recent work by Kim et al. (2016) where in addition to selecting prototypes, we also associate non-negative weights which are indicative of their importance. This extension provides a single coherent framework under which both prototypes and criticisms (i.e. outliers) can be found. Furthermore, our framework works for any symmetric positive definite kernel thus addressing one of the key open questions laid out in Kim et al. (2016). By establishing that our objective function enjoys a key property of that of weak submodularity, we present a fast ProtoDash algorithm and also derive approximation guarantees for the same. We demonstrate the efficacy of our method on diverse domains such as retail, digit recognition (MNIST) and on publicly available 40 health questionnaires obtained from the Center for Disease Control (CDC) website maintained by the US Dept. of Health. We validate the results quantitatively as well as qualitatively based on expert feedback and recently published scientific studies on public health, thus showcasing the power of our technique in providing actionability (for retail), utility (for MNIST), and insight (on CDC datasets), which arguably are the hallmarks of an effective interpretable machine learning method.

Drawing an inspiration from behavioral studies of human decision making, we propose here a general parametric framework for a reinforcement learning problem, which extends the standard Q-learning approach to incorporate a two-stream framework of reward processing with biases biologically associated with several neurological and psychiatric conditions, including Parkinson's and Alzheimer's diseases, attention-deficit/hyperactivity disorder (ADHD), addiction, and chronic pain. For AI community, the development of agents that react differently to different types of rewards can enable us to understand a wide spectrum of multi-agent interactions in complex real-world socioeconomic systems. Moreover, from the behavioral modeling perspective, our parametric framework can be viewed as a first step towards a unifying computational model capturing reward processing abnormalities across multiple mental conditions and user preferences in long-term recommendation systems.

ALS is a fatal neurodegenerative disease with no cure. Experts typically measure disease progression via the ALSFRS-R score, which includes measurements of various abilities known to decline. We propose instead the use of speech analysis as a proxy for ALS progression. This technique enables 1) frequent non-invasive, inexpensive, longitudinal analysis, 2) analysis of data recorded in the wild, and 3) creation of an extensive ALS databank for future analysis. Patients and trained medical professionals need not be co-located, enabling more frequent monitoring of more patients from the convenience of their own homes. The goals of this study are the identification of acoustic speech features in naturalistic contexts which characterize disease progression and development of machine models which can recognize the presence and severity of the disease. We evaluated subjects from the Prize4Life Israel dataset, using a variety of frequency, spectral, and voice quality features. The dataset was generated using the ALS Mobile Analyzer, a cell- phone app that collects data regarding disease progress using a self-reported ALSFRS-R questionnaire and several active tasks that measure speech and motor skills. Classification via leave- five-subjects-out cross-validation resulted in an accuracy rate of 79% (61% chance) for males and 83% (52% chance) for females.

We present a theory of decision-making in the presence of multiple choices that departs from traditional approaches by explicitly incorporating entropic barriers in a stochastic search process. We analyze response time data from an on-line repository of 15 million blitz chess games, and show that our model fits not just the mean and variance, but the entire response time distribution (over several response-time orders of magnitude) at every stage of the game. We apply the model to show that (a) higher cognitive expertise corresponds to the exploration of more complex solution spaces, and (b) reaction times of users at an on-line buying website can be similarly explained. Our model can be seen as a synergy between diffusion models used to model simple two-choice decision-making and planning agents in complex problem solving.

The diagnosis of serious mental health conditions such as schizophrenia is based on the judgment of clinicians whose training takes many years and cannot be easily formalized into objective measures.However, clinical research suggests there are disturbances in aspects of the language use of patients with schizophrenia, which opens a door for the use of NLP tools in schizophrenia diagnosis and prognosis.Using metaphor-identification and sentiment-analysis algorithms to automatically generate features, we create a classifier that, with high accuracy, can predict which patients will develop (or currently suffer from) schizophrenia.To our knowledge, this study is the first to demonstrate the utility of automated metaphor identification algorithms for detection or prediction of disease.

We consider a novel formulation of the multi-armed bandit model, which we call the contextual bandit with restricted context, where only a limited number of features can be accessed by the learner at every iteration. This novel formulation is motivated by different online problems arising in clinical trials, recommender systems and attention modeling.Herein, we adapt the standard multi-armed bandit algorithm known as Thompson Sampling to take advantage of our restricted context setting, and propose two novel algorithms, called the Thompson Sampling with Restricted Context (TSRC) and the Windows Thompson Sampling with Restricted Context (WTSRC), for handling stationary and nonstationary environments, respectively. Our empirical results demonstrate advantages of the proposed approaches on several real-life datasets.

We address the problem of online model adaptation when learning representations from non-stationary data streams. Specifically, we focus here on online dictionary learning (i.e. sparse linear autoencoder), and propose a simple but effective online model selection approach involving “birth” (addition) and “death” (removal) of hidden units representing dictionary elements, in response to changing inputs; we draw inspiration from the adult neurogenesis phenomenon in the dentate gyrus of the hippocampus, known to be associated with better adaptation to new environments. Empirical evaluation on real-life datasets (images and text), as well as on synthetic data, demonstrates that the proposed approach can considerably outperform the state-of-art non-adaptive online sparse coding of [Mairal et al., 2009] in the presence of non-stationary data. Moreover, we identify certain data- and model properties associated with such improvements.

Speech data has the potential to become a powerful tool to provide quantitative information about emotion beyond that achieved by subjective assessments. Based on this concept, we investigate the use of speech to identify effects in subjects under the influence of two different drugs: Oxytocin (OT) and 3, 4-methylenedioxymethamphetamine (MDMA), also known as ecstasy. We extract a set of informative phonological features that can characterize emotion. Then, we perform classification to detect if the subject is under the influence of a drug. Our best results show low error rates of 13% and 17% for the subject classification of OT and MDMA vs. placebo, respectively. We also analyze the performance of the features to differentiate the two levels of MDMA doses, obtaining an error rate of 19%. The results indicate that subtle emotional changes can be detected in the context of drug use.

The external world consists of objects that stimulate multiple sensory pathways simultaneously, such as auditory, visual and touch. Our brains receive and process these sensory streams to arrive at a coherent internal representation of the world. Though much attention has been paid to these streams individually, their integration is comparatively less well understood. In this paper we propose the principle of sparse spatio-temporal encoding as a foundation to build a framework for multi-sensory integration. We derive the dynamics that govern a network of oscillatory units that achieves phase synchronization, and is capable of binding related attributes of objects. We simulate objects that produce simultaneous visual and auditory input activations. We demonstrate that our system can bind features in both these sensory modalities. We examine the effect of varying a tuning function that governs the ability of the units to synchronize, and show that by broadening this function we reduce the ability of the network to disambiguate mixtures of objects. Thus, our model offers the potential to study brain disorders such as autism, which may arise from a disruption of synchronization.

Psychological studies have investigated limits in the capacity to simultaneously store multiple objects in working memory, which turns out to be approximately four. In this paper we examine the existence and origin of such a capacity limit in the sensory encoding stage, where synchronous activity can be considered to group related features of a common object. We develop a model of an object recognition network using oscillatory elements that can achieve phase synchronization. Using simulations based on this network, we show that distinct phases of oscillation can be used to label combinations of objects presented simultaneously. This allows the network to separate mixtures of objects, and identify the input elements that belong to each object. We demonstrate that there is a limit of four objects that can be separated from a mixture. Further studies are required to generalize this result by varying the size of the network and the number of objects used for training. We also show that by narrowing a tuning function that governs the dynamics of the system, we can achieve higher separation accuracy. However, this comes at the cost of utilizing a higher number of iterations over which the system settles and learns its synaptic weights. We lay down a framework for the quantitative modeling of factors affecting capacity limits, which has the potential to advance our understanding of sensory representation, attention, working memory and multitasking.

While the static magnitude of thermal pain perception has been shown to follow a power-law function of the temperature, its dynamical features have been largely overlooked. Due to the slow temporal experience of pain, multiple studies now show that the time evolution of its magnitude can be captured with continuous online ratings. Here we use such ratings to model quantitatively the temporal dynamics of thermal pain perception. We show that a differential equation captures the details of the temporal evolution in pain ratings in individual subjects for different stimulus pattern complexities, and also demonstrates strong predictive power to infer pain ratings, including readouts based only on brain functional images.

We model neural dynamical behavior during object perception using the principle of sparse coding in multilayer oscillatory networks. The network model consists of units with amplitude and phase variables, and allows the propagation of higher-level information to lower levels via feedback connections. We show that this model can replicate findings in the neuroscience literature, where measurements have shown that neurons in lower level visual areas respond in a delayed fashion to missing contours of whole objects. We contrast the behavior of feedback connections with that of lateral connections by selectively disabling these in our model to examine their contributions to object perception. This paper successfully extends the previously reported capabilities of oscillatory networks by applying them to model perceptual tasks.

The visual cortex contains well-ordered maps for visual features such as orientation. Existing computational approaches have demonstrated that biologically realistic global organization can be obtained. However, specific local properties such as variability in orientation tuning have not been thoroughly investigated. The purpose of this paper is to provide a computationally grounded interpretation of the evolution of topographic properties of cortical maps. We use a modified Kohonen approach to self-organization, and show that the orientation tuning of cortical units varies such that orientation selectivity is weaker at pinwheel centers than linear zones. This finding impacts a debate in neuroscience, where researchers are trying to resolve whether or not orientation tuning is sharp everywhere. We also examined a null-hypothesis, which modified the behavior of cortical units so that they are free to choose their orientation preference, but keep their orientation tuning constant. Our approach offers a mechanistic interpretation for how local learning algorithms can shape global organization. Our results show that a spatially differentiated orientation tuning map is intrinsic to a spatial learning process. Furthermore, constraining the orientation tuning of cortical units to be constant results in unacceptable map formation.

The relationship between structure and function is of central importance in neuroscience. Computational modeling techniques can play a crucial role in exploring this relationship. Neuroscientists have revealed an interesting patterning in the connectivity of visual cortical areas, where the receptive field sizes for feed-forward, lateral and feedback connections are monotonically increasing and these roughly double. In this paper, we use a computational modeling approach to understand the behavior of the visual system, and show that this observed connectivity pattern can be explained via a maximization of functional metrics based on separation and segmentation accuracies. We use an optimization function based on sparse spatiotemporal encoding and faithfulness of representation to derive the dynamical behavior of a multi-layer network of oscillatory units. The network behavior can be quantified in terms of its ability to separate and segment mixtures of inputs. We vary the topological connectivity between different layers of the simulated visual network and study its effect on performance in terms of separation and segmentation accuracy. We demonstrate that the best performance occurs when the topology of the simulated system is similar to that in the primate visual cortex where the receptive field sizes of feedforward, lateral and feedback connections are monotonically increasing. This explanation of the functional significance of topological connectivity provides a new perspective for the understanding of cortical function.

Schizophrenia is a complex psychiatric disorder that has eluded a characterization in terms of local abnormalities of brain activity, and is hypothesized to affect the collective, ``emergent working of the brain. We propose a novel data-driven approach to capture emergent features using functional brain networks [Eguiluzet al] extracted from fMRI data, and demonstrate its advantage over traditional region-of-interest (ROI) and local, task-specific linear activation analyzes. Our results suggest that schizophrenia is indeed associated with disruption of global, emergent brain properties related to its functioning as a network, which cannot be explained by alteration of local activation patterns. Moreover, further exploitation of interactions by sparse Markov Random Field classifiers shows clear gain over linear methods, such as Gaussian Naive Bayes and SVM, allowing to reach 86% accuracy (over 50% baseline - random guess), which is quite remarkable given that it is based on a single fMRI experiment using a simple auditory task.

We propose a family of supervised dimensionality reduction (SDR) algorithms that combine feature extraction (dimensionality reduction) with learning a predictive model in a unified optimization framework, using data- and class-appropriate generalized linear models (GLMs), and handling both classification and regression problems. Our approach uses simple closed-form update rules and is provably convergent. Promising empirical results are demonstrated on a variety of high-dimensional datasets.

In earlier work, we derived the dynamical behavior of a network of oscillatory units described by the amplitude and phase of oscillations. The dynamics were derived from an objective function that rewards both the faithfulness and the sparseness of representation. After unsupervised learning, the network is capable of separating mixtures of inputs, and also segmenting the inputs into components that most contribute to the classification of a given input object. In the current paper, we extend our analysis to multi-layer networks, and demonstrate that the dynamical equations derived earlier can be successfully applied to multi-layer networks. The topological connectivity between the different layers are derived from biological observations in primate visual cortex, and consist of receptive fields that are topographically mapped between layers. We explore the role of feedback connections, and show that increasing the diffusivity of feedback connections significantly improves segmentation performance, but does not affect separation performance.

In this paper, we present a novel network to separate mixtures of inputs that have been previously learned. A significant capability of the network is that it segments the components of each input object that most contribute to its classification. The network consists of amplitude-phase units that can synchronize their dynamics, so that separation is determined by the amplitude of units in an output layer, and segmentation by phase similarity between input and output layer units. Learning is unsupervised and based on a Hebbian update, and the architecture is very simple. Moreover, efficient segmentation can be achieved even when there is considerable superposition of the inputs. The network dynamics are derived from an objective function that rewards sparse coding in the generalized amplitude-phase variables. We argue that this objective function can provide a possible formal interpretation of the binding problem and that the implementation of the network architecture and dynamics is biologically plausible.

Information about various visual attributes, such as orientation or color, is represented in the primate striate cortex (VI) before reaching extrastriate cortices. Previous studies, using single-or multi-unit recordings, have reached conflicting conclusions regarding the question of whether these different attributes are represented by separate populations of neurons in VI. We addressed this question with imaging of intrinsic optical signals from macaque VI. The animals viewed either full field (uniform) color stimuli or drifting black and white gratings of various orientations. We used the Support Vector Machine (SVM) algorithm to determine the relative amount of information about a given attribute at each cortical location. We found that each single-trial image contained significant amounts of information about color or orientation. The information about either attribute was distributed unevenly across VI. Color information was concentrated in small regions of high information density, surrounded by large regions of low information density, while orientation information showed the opposite pattern. Our results are consistent with the hypothesis that color and orientation are represented in different compartments in VI.

We present a novel network of oscillatory units, whose behavior is described by the amplitude and phase of oscillations. We derive the network dynamics from an objective function that rewards both the faithfulness and the sparseness of representation. The resulting network architecture is simple, and the dynamics are straightforward to interpret. This network functions in an unsupervised manner, and is able to form unique representations for a set of inputs. Once the set of inputs is learnt, the network can deconvolve mixtures of inputs. A significant capability of the network is that it segments its inputs into components that most contribute to the classification of a given input object. Network units exhibit synchronization through phase locking after an initial settling period. The behavior of deconvolution is determined by the amplitude of units in an output layer, while segmentation is simultaneously determined through phase similarity between the input and output layer units. Thus, the network exhibits the binding of specific input units and the output units that represent a classification of these input units by means of phase similarity.