Search

Objective: Most current wearable tonic-clonic seizure (TCS) detection systems are based on extra-cerebral signals, such as electromyography (EMG) or accelerometry (ACC). Although many of these devices show good sensitivity in seizure detection, their false positive rates (FPR) are still relatively high. Wearable EEG may improve performance; however, studies investigating this remain scarce. This paper aims 1) to investigate the possibility of detecting TCSs with a behind-the-ear, two-channel wearable EEG, and 2) to evaluate the added value of wearable EEG to other non-EEG modalities in multimodal TCS detection. Method: We included 27 participants with a total of 44 TCSs from the European multicenter study SeizeIT2. The multimodal wearable detection system Sensor Dot (Byteflies) was used to measure two-channel, behind-the-ear EEG, EMG, electrocardiography (ECG), ACC and gyroscope (GYR). First, we evaluated automatic unimodal detection of TCSs, using performance metrics such as sensitivity, precision, FPR and F1-score. Secondly, we fused the different modalities and again assessed performance. Algorithm-labeled segments were then provided to a neurologist and a wearable data expert, who reviewed and annotated the true positive TCSs, and discarded false positives (FPs). Results: Wearable EEG outperformed the other modalities in unimodal TCS detection by achieving a sensitivity of 100.0% and a FPR of 10.3/24h (compared to 97.7% sensitivity and 30.9/24h FPR for EMG; 95.5% sensitivity and 13.9 FPR for ACC). The combination of wearable EEG and EMG achieved overall the most clinically useful performance in offline TCS detection with a sensitivity of 97.7%, a FPR of 0.4/24 h, a precision of 43.0%, and a F1-score of 59.7%. Subsequent visual review of the automated detections resulted in maximal sensitivity and zero FPs.

On August 9-10, 2023, a workshop was convened at the Pacific Northwest National Laboratory (PNNL) in Richland, WA that brought together a group of internationally recognized experts in metabolomics, natural products discovery, chemical ecology, chemical and biological threat assessment, cheminformatics, computational chemistry, cloud computing, artificial intelligence, and novel technology development. These experts were invited to assess the value and feasibility of a grand-scale project to create new technologies that would allow the identification and quantification of all small molecules, or to decode the molecular universe. The Decoding the Molecular Universe project would extend and complement the success of the Human Genome Project by developing new capabilities and technologies to measure small molecules (defined as non-protein, non-polymer molecules less than 1500 Daltons) of any origin and generated in biological systems or produced abiotically. Workshop attendees 1) explored what new understanding of biological and environmental systems could be revealed through the lens of small molecules; 2) characterized the similarities in current needs and technical challenges between each science or mission area for unambiguous and comprehensive determination of the composition and quantities of small molecules of any sample; 3) determined the extent to which technologies or methods currently exist for unambiguously and comprehensively determining the small molecule composition of any sample and in a reasonable time; and 4) identified the attributes of the ideal technology or approach for universal small molecule measurement and identification. The workshop concluded with a discussion of how a project of this scale could be undertaken, possible thrusts for the project, early proof-of-principle applications, and similar efforts upon which the project could be modeled.

State of the art reinforcement learning has enabled training agents on tasks of ever increasing complexity. However, the current paradigm tends to favor training agents from scratch on every new task or on collections of tasks with a view towards generalizing to novel task configurations. The former suffers from poor data efficiency while the latter is difficult when test tasks are out-of-distribution. Agents that can effectively transfer their knowledge about the world pose a potential solution to these issues. In this paper, we investigate transfer learning in the context of model-based agents. Specifically, we aim to understand when exactly environment models have an advantage and why. We find that a model-based approach outperforms controlled model-free baselines for transfer learning. Through ablations, we show that both the policy and dynamics model learnt through exploration matter for successful transfer. We demonstrate our results across three domains which vary in their requirements for transfer: in-distribution procedural (Crafter), in-distribution identical (RoboDesk), and out-of-distribution (Meta-World). Our results show that intrinsic exploration combined with environment models present a viable direction towards agents that are self-supervised and able to generalize to novel reward functions.

We introduce a challenging decision-making task that we call active acquisition for multimodal temporal data (A2MT). In many real-world scenarios, input features are not readily available at test time and must instead be acquired at significant cost. With A2MT, we aim to learn agents that actively select which modalities of an input to acquire, trading off acquisition cost and predictive performance. A2MT extends a previous task called active feature acquisition to temporal decision making about high-dimensional inputs. We propose a method based on the Perceiver IO architecture to address A2MT in practice. Our agents are able to solve a novel synthetic scenario requiring practically relevant cross-modal reasoning skills. On two large-scale, real-world datasets, Kinetics-700 and AudioSet, our agents successfully learn cost-reactive acquisition behavior. However, an ablation reveals they are unable to learn adaptive acquisition strategies, emphasizing the difficulty of the task even for state-of-the-art models. Applications of A2MT may be impactful in domains like medicine, robotics, or finance, where modalities differ in acquisition cost and informativeness., Comment: Published in Transactions on Machine Learning Research. Previous version accepted to Foundation Models for Decision Making Workshop at NeurIPS 2022

Using a model of the environment and a value function, an agent can construct many estimates of a state's value, by unrolling the model for different lengths and bootstrapping with its value function. Our key insight is that one can treat this set of value estimates as a type of ensemble, which we call an \emph{implicit value ensemble} (IVE). Consequently, the discrepancy between these estimates can be used as a proxy for the agent's epistemic uncertainty; we term this signal \emph{model-value inconsistency} or \emph{self-inconsistency} for short. Unlike prior work which estimates uncertainty by training an ensemble of many models and/or value functions, this approach requires only the single model and value function which are already being learned in most model-based reinforcement learning algorithms. We provide empirical evidence in both tabular and function approximation settings from pixels that self-inconsistency is useful (i) as a signal for exploration, (ii) for acting safely under distribution shifts, and (iii) for robustifying value-based planning with a learned model., Comment: The first three authors contributed equally. Accepted at ICML 2022

One of the key promises of model-based reinforcement learning is the ability to generalize using an internal model of the world to make predictions in novel environments and tasks. However, the generalization ability of model-based agents is not well understood because existing work has focused on model-free agents when benchmarking generalization. Here, we explicitly measure the generalization ability of model-based agents in comparison to their model-free counterparts. We focus our analysis on MuZero (Schrittwieser et al., 2020), a powerful model-based agent, and evaluate its performance on both procedural and task generalization. We identify three factors of procedural generalization -- planning, self-supervised representation learning, and procedural data diversity -- and show that by combining these techniques, we achieve state-of-the art generalization performance and data efficiency on Procgen (Cobbe et al., 2019). However, we find that these factors do not always provide the same benefits for the task generalization benchmarks in Meta-World (Yu et al., 2019), indicating that transfer remains a challenge and may require different approaches than procedural generalization. Overall, we suggest that building generalizable agents requires moving beyond the single-task, model-free paradigm and towards self-supervised model-based agents that are trained in rich, procedural, multi-task environments.

Transcriptomics response of SK-N-AS cells to methamidophos (an acetylcholine esterase inhibitor) exposure was measured at 10 time points between 0.5 and 48 h. The data was analyzed using a combination of traditional statistical methods and novel machine learning algorithms for detecting anomalous behavior and infer causal relations between time profiles. We identified several processes that appeared to be upregulated in cells treated with methamidophos including: unfolded protein response, response to cAMP, calcium ion response, and cell-cell signaling. The data confirmed the expected consequence of acetylcholine buildup. In addition, transcripts with potentially key roles were identified and causal networks relating these transcripts were inferred using two different computational methods: Siamese convolutional networks and time warp causal inference. Two types of anomaly detection algorithms, one based on Autoencoders and the other one based on Generative Adversarial Networks (GANs), were applied to narrow down the set of relevant transcripts.

In spite of the rapidly increasing number of applications of machine learning in various domains, a principled and systematic approach to the incorporation of domain knowledge in the engineering process is still lacking and ad hoc solutions that are difficult to validate are still the norm in practice, which is of growing concern not only in mission-critical applications. In this note, we introduce Probabilistic Approximate Logic (PALO) as a logic based on the notion of mean approximate probability to overcome conceptual and computational difficulties inherent to strictly probabilistic logics. The logic is approximate in several dimensions. Logical independence assumptions are used to obtain approximate probabilities, but by averaging over many instances of formulas a useful estimate of mean probability with known confidence can usually be obtained. To enable efficient computational inference, the logic has a continuous semantics that reflects only a subset of the structural properties of classical logic, but this imprecision can be partly compensated by richer theories obtained by classical inference or other means. Computational inference, which refers to the construction of models and validation of logical properties, is based on Stochastic Gradient Descent (SGD) and Markov Chain Monte Carlo (MCMC) techniques and hence another dimension where approximations are involved. We also present the Logical Imagination Engine (LIME), a prototypical implementation of PALO based on TensorFlow. Albeit not limited to the biological domain, we illustrate its operation in a quite substantial bioinformatics machine learning application concerned with network synthesis and analysis in a recent DARPA project.

Animals need to devise strategies to maximize returns while interacting with their environment based on incoming noisy sensory observations. Task-relevant states, such as the agent's location within an environment or the presence of a predator, are often not directly observable but must be inferred using available sensory information. Successor representations (SR) have been proposed as a middle-ground between model-based and model-free reinforcement learning strategies, allowing for fast value computation and rapid adaptation to changes in the reward function or goal locations. Indeed, recent studies suggest that features of neural responses are consistent with the SR framework. However, it is not clear how such representations might be learned and computed in partially observed, noisy environments. Here, we introduce a neurally plausible model using distributional successor features, which builds on the distributed distributional code for the representation and computation of uncertainty, and which allows for efficient value function computation in partially observed environments via the successor representation. We show that distributional successor features can support reinforcement learning in noisy environments in which direct learning of successful policies is infeasible.

There is an abundance of complex dynamic systems that are critical to our daily lives and our society but that are hardly understood, and even with today's possibilities to sense and collect large amounts of experimental data, they are so complex and continuously evolving that it is unlikely that their dynamics will ever be understood in full detail. Nevertheless, through computational tools we can try to make the best possible use of the current technologies and available data. We believe that the most useful models will have to take into account the imbalance between system complexity and available data in the context of limited knowledge or multiple hypotheses. The complex system of biological cells is a prime example of such a system that is studied in systems biology and has motivated the methods presented in this paper. They were developed as part of the DARPA Rapid Threat Assessment (RTA) program, which is concerned with understanding of the mechanism of action (MoA) of toxins or drugs affecting human cells. Using a combination of Gaussian processes and abstract network modeling, we present three fundamentally different machine-learning-based approaches to learn causal relations and synthesize causal networks from high-dimensional time series data. While other types of data are available and have been analyzed and integrated in our RTA work, we focus on transcriptomics (that is gene expression) data obtained from high-throughput microarray experiments in this paper to illustrate capabilities and limitations of our algorithms. Our algorithms make different but overall relatively few biological assumptions, so that they are applicable to other types of biological data and potentially even to other complex systems that exhibit high dimensionality but are not of biological nature.

The study of brain networks, including derived from functional neuroimaging data, attracts broad interest and represents a rapidly growing interdisciplinary field. Comparing networks of healthy volunteers with those of patients can potentially offer new, quantitative diagnostic methods, and a framework for better understanding brain and mind disorders. We explore resting state fMRI data through network measures, and demonstrate that not only is there a distinctive network architecture in the healthy brain that is disrupted in schizophrenia, but also that both networks respond to medication. We construct networks representing 15 healthy individuals and 12 schizophrenia patients (males and females), all of whom are administered three drug treatments: (i) a placebo; and two antipsychotic medications (ii) aripiprazole and; (iii) sulpiride. We first reproduce the established finding that brain networks of schizophrenia patients exhibit increased efficiency and reduced clustering compared to controls. Our data then reveals that the antipsychotic medications mitigate this effect, shifting the metrics towards those observed in healthy volunteers, with a marked difference in efficacy between the two drugs. Additionally, we find that aripiprazole considerably alters the network statistics of healthy controls. Using a test of cognitive ability, we establish that aripiprazole also adversely affects their performance. This provides evidence that changes to macroscopic brain network architecture result in measurable behavioural differences. This is the first time different medications have been assessed in this way. Our results lay the groundwork for an objective methodology with which to calculate and compare the efficacy of different treatments of mind and brain disorders., Comment: 18 pages, 2 figures, 3 tables

Control is essential to the functioning of any neural system. Indeed, under healthy conditions the brain must be able to continuously maintain a tight functional control between the system's inputs and outputs. One may therefore hypothesise that the brain's wiring is predetermined by the need to maintain control across multiple scales, maintaining the stability of key internal variables, and producing behaviour in response to environmental cues. Recent advances in network control have offered a powerful mathematical framework to explore the structure-function relationship in complex biological, social, and technological networks, and are beginning to yield important and precise insights for neuronal systems. The network control paradigm promises a predictive, quantitative framework to unite the distinct datasets necessary to fully describe a nervous system, and provide mechanistic explanations for the observed structure and function relationships. Here, we provide a thorough review of the network control framework as applied to C. elegans, in the style of a FAQ. We present the theoretical, computational, and experimental aspects of network control, and discuss its current capabilities and limitations, together with the next likely advances and improvements. We further present the Python code to enable exploration of control principles in a manner specific to this prototypical organism., Comment: 19 pages, 5 figures, 1 table

We introduce a new approach to learning in hierarchical latent-variable generative models called the "distributed distributional code Helmholtz machine", which emphasises flexibility and accuracy in the inferential process. In common with the original Helmholtz machine and later variational autoencoder algorithms (but unlike adverserial methods) our approach learns an explicit inference or "recognition" model to approximate the posterior distribution over the latent variables. Unlike in these earlier methods, the posterior representation is not limited to a narrow tractable parameterised form (nor is it represented by samples). To train the generative and recognition models we develop an extended wake-sleep algorithm inspired by the original Helmholtz Machine. This makes it possible to learn hierarchical latent models with both discrete and continuous variables, where an accurate posterior representation is essential. We demonstrate that the new algorithm outperforms current state-of-the-art methods on synthetic, natural image patch and the MNIST data sets.

Graph theoretical analysis of the community structure of networks attempts to identify the communities (or modules) to which each node affiliates. However, this is in most cases an ill-posed problem, as the affiliation of a node to a single community is often ambiguous. Previous solutions have attempted to identify all of the communities to which each node affiliates. Instead of taking this approach, we introduce versatility, $V$, as a novel metric of nodal affiliation: $V \sim 0$ means that a node is consistently assigned to a specific community; $V \gg 0$ means it is inconsistently assigned to different communities. Versatility works in conjunction with existing community detection algorithms, and it satisfies many theoretically desirable properties in idealised networks designed to maximise ambiguity of modular decomposition. The local minima of global mean versatility identified the resolution parameters of a hierarchical community detection algorithm that least ambiguously decomposed the community structure of a social (karate club) network and the mouse brain connectome. Our results suggest that nodal versatility is useful in quantifying the inherent ambiguity of modular decomposition.

Connectomics has focused primarily on the mapping of synaptic links in the brain; yet it is well established that extrasynaptic volume transmission, especially via monoamines and neuropeptides, is also critical to brain function. Here we present a draft monoamine connectome, along with a partial neuropeptide connectome, for the nematode C. elegans, based on new and published expression data for biosynthetic genes and receptors. Thus, the neuronal connectome can be represented as a multiplex network, with synaptic, gap junction, and neuromodulatory layers representing alternative modes of interneuronal interaction and with distinct network structures. In particular, the monoamine network exhibits novel topological properties, with a highly disassortative star-like structure and a rich-club of interconnected broadcasting hubs. Despite the low degree of overlap between layers, we find highly significant modes of interaction, pinpointing network locations and multilink motifs where aminergic and neuropeptide signalling modulate synaptic activity. The multilayer connectome of C. elegans represents a clear exemplar of a biological multiplex network and provides a prototype for understanding how extrasynaptic signalling can be integrated into the wired circuitry in larger brains.