Search

This work provides a theoretical framework for assessing the generalization error of graph classification tasks via graph neural networks in the over-parameterized regime, where the number of parameters surpasses the quantity of data points. We explore two widely utilized types of graph neural networks: graph convolutional neural networks and message passing graph neural networks. Prior to this study, existing bounds on the generalization error in the over-parametrized regime were uninformative, limiting our understanding of over-parameterized network performance. Our novel approach involves deriving upper bounds within the mean-field regime for evaluating the generalization error of these graph neural networks. We establish upper bounds with a convergence rate of $O(1/n)$, where $n$ is the number of graph samples. These upper bounds offer a theoretical assurance of the networks' performance on unseen data in the challenging over-parameterized regime and overall contribute to our understanding of their performance., Comment: 43 pages

In recent years, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) has conducted a program to re-evaluate the safety of natural flavor complexes (NFCs) used as flavor ingredients. This publication, twelfth in the series, details the re-evaluation of NFCs whose constituent profiles are characterized by alicyclic or linear ketones. In its re-evaluation, the Expert Panel applies a scientific constituent-based procedure for the safety evaluation of NFCs in commerce using a congeneric group approach. Estimated intakes of each congeneric group of the NFC are evaluated using the well-established and conservative Threshold of Toxicological Concern (TTC) approach. In addition, studies on the toxicity and genotoxicity of members of the congeneric groups and the NFCs under evaluation are reviewed. The scope of the safety evaluation of the NFCs contained herein does not include added use in dietary supplements or any products other than food. Thirteen (13) NFCs derived from the Boronia, Cinnamomum, Thuja, Ruta, Salvia, Tagetes, Hyssopus, Iris, Perilla and Artemisia genera are affirmed as generally recognized as safe (GRAS) under conditions of their intended use as flavor ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

Samuel A Cohen,1,2 Ann Caroline Fisher,1,2 Suzann Pershing1– 3 1Department of Ophthalmology, Stanford University School of Medicine, Stanford, CA, USA; 2Byers Eye Institute at Stanford, Stanford, CA, USA; 3VA Palo Alto Health Care System, Palo Alto, CA, USACorrespondence: Suzann Pershing, Department of Ophthalmology, Stanford University, 2452 Watson Court, Stanford, CA, USA, 94303, Tel +1 650 723-6995, Email pershing@stanford.eduPurpose: To assess the readability and accountability of online patient education materials related to glaucoma diagnosis and treatment.Methods: We conducted a Google search for 10 search terms related to glaucoma diagnosis and 10 search terms related to glaucoma treatment. For each search term, the first 10 patient education websites populated after Google search were assessed for readability and accountability. Readability was assessed using five validated measures: Flesch Reading Ease (FRE), Gunning Fog Index (GFI), Flesch-Kincaid Grade Level (FKGL), Simple Measure of Gobbledygook (SMOG), and New Dale-Chall (NDC). Accountability was assessed using the Journal of the American Medical Association (JAMA) benchmarks. The source of information for each article analyzed was recorded.Results: Of the 200 total websites analyzed, only 11% were written at or below the recommended 6th grade reading level. The average FRE and grade level for 100 glaucoma diagnosis-related articles were 42.02 ± 1.08 and 10.53 ± 1.30, respectively. The average FRE and grade level for 100 glaucoma treatment-related articles were 43.86 ± 1.01 and 11.29 ± 1.54, respectively. Crowdsourced articles were written at the highest average grade level (12.32 ± 0.78), followed by articles written by private practice/independent users (11.22 ± 1.74), national organizations (10.92 ± 1.24), and educational institutions (10.33 ± 1.35). Websites averaged 1.12 ± 1.15 of 4 JAMA accountability metrics.Conclusion: Despite wide variation in the readability and accountability of o

The Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) applies its procedure for the safety evaluation of natural flavor complexes (NFCs) to re-evaluate the safety of Asafetida Oil (Ferula assa-foetida L.) FEMA 2108, Garlic Oil (Allium sativum L.) FEMA 2503 and Onion Oil (Allium cepa L.) FEMA 2817 for use as flavoring in food. This safety evaluation is part of a series of evaluations of NFCs for use as flavoring ingredients conducted by the Expert Panel that applies a scientific procedure published in 2005 and updated in 2018. Using a group approach that relies on a complete chemical characterization of the NFC intended for commerce, the constituents of each NFC are organized into well-defined congeneric groups and the estimated intake of each constituent congeneric group is evaluated using the conservative threshold of toxicological concern (TTC) concept. Data on the metabolism, genotoxic potential and toxicology for each constituent congeneric group are reviewed as well as studies on each NFC. Based on the safety evaluation, Asafetida Oil (Ferula assa-foetida L.), Garlic Oil (Allium sativum L.) and Onion Oil (Allium cepa L.) were affirmed as generally recognized as safe (GRASa) under their conditions of intended use as flavor ingredients.

The FEMA Expert Panel program to re-evaluate the safety of natural flavor complexes (NFCs) used as flavoring ingredients in food has resulted in the publication of an updated constituent-based procedure as well as publications on the safety evaluation of many botanical-derived NFCs. This publication, ninth in the series and related to the ninth publication, describes the affirmation of the generally recognized as safe (GRAS) status for NFCs with propenylhydroxybenzene and allylalkoxybenzene constituents under their conditions of intended use as flavoring ingredients added to food. The Panel's procedure applies the threshold of toxicological concern (TTC) concept and evaluates relevant data on absorption, metabolism, genotoxic potential and toxicology for the NFCs themselves and their respective constituent congeneric groups. For NFCs containing allylalkoxybenzene constituent(s) with suspected genotoxic potential, the estimated intake of the individual constituent is compared to the TTC for compounds with structural alerts for genotoxicity and if exceeded, a margin of exposure is calculated using BMDL10 values derived from benchmark dose analyses using Bayesian model averaging, as presented in the tenth article of the series. Safety evaluations for NFCs derived from allspice, anise seed, star anise, sweet fennel seed and pimento leaves were conducted and their GRAS status was affirmed for use as flavoring ingredients. The scope of the safety evaluation contained herein does not include added use in dietary supplements or any products other than food

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a program for the re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients. This publication, eleventh in the series, evaluates the safety of NFCs characterized by primary alcohol, aldehyde, carboxylic acid, ester and lactone constituents derived from terpenoid biosynthetic pathways and/or lipid metabolism. The scientific-based evaluation procedure published in 2005 and updated in 2018 that relies on a complete constituent characterization of the NFC and organization of the constituents into congeneric groups. The safety of the NFCs is evaluated using the threshold of toxicological concern (TTC) concept in addition to data on estimated intake, metabolism and toxicology of members of the congeneric groups and for the NFC under evaluation. The scope of the safety evaluation does not include added use in dietary supplements or any products other than food. Twenty-three NFCs, derived from the Hibiscus, Melissa, Ricinus, Anthemis, Matricaria, Cymbopogon, Saussurea, Spartium, Pelargonium, Levisticum, Rosa, Santalum, Viola, Cryptocarya and Litsea genera were affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a program for the re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavoring ingredients in food. In this publication, tenth in the series, NFCs containing a high percentage of at least one naturally occurring allylalkoxybenzene constituent with a suspected concern for genotoxicity and/or carcinogenicity are evaluated. In a related paper, ninth in the series, NFCs containing anethole and/or eugenol and relatively low percentages of these allylalkoxybenzenes are evaluated. The Panel applies the threshold of toxicological concern (TTC) concept and evaluates relevant toxicology data on the NFCs and their respective constituent congeneric groups. For NFCs containing allylalkoxybenzene constituent(s), the estimated intake of the constituent is compared to the TTC for compounds with structural alerts for genotoxicity and when exceeded, a margin of exposure (MOE) is calculated. BMDL10 values are derived from benchmark dose analyses using Bayesian model averaging for safrole, estragole and methyl eugenol using EPA's BMDS software version 3.2. BMDL10 values for myristicin, elemicin and parsley apiole were estimated by read-across using relative potency factors. Margins of safety for each constituent congeneric group and MOEs for each allylalkoxybenzene constituent for each NFC were determined that indicate no safety concern. The scope of the safety evaluation contained herein does not include added use in dietary supplements or any products other than food. Ten NFCs, derived from basil, estragon (tarragon), mace, nutmeg, parsley and Canadian snakeroot were determined or affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

Numerically solving high-dimensional partial differential equations (PDEs) is a major challenge. Conventional methods, such as finite difference methods, are unable to solve high-dimensional PDEs due to the curse-of-dimensionality. A variety of deep learning methods have been recently developed to try and solve high-dimensional PDEs by approximating the solution using a neural network. In this paper, we prove global convergence for one of the commonly-used deep learning algorithms for solving PDEs, the Deep Galerkin Method (DGM). DGM trains a neural network approximator to solve the PDE using stochastic gradient descent. We prove that, as the number of hidden units in the single-layer network goes to infinity (i.e., in the ``wide network limit"), the trained neural network converges to the solution of an infinite-dimensional linear ordinary differential equation (ODE). The PDE residual of the limiting approximator converges to zero as the training time $\rightarrow \infty$. Under mild assumptions, this convergence also implies that the neural network approximator converges to the solution of the PDE. A closely related class of deep learning methods for PDEs is Physics Informed Neural Networks (PINNs). Using the same mathematical techniques, we can prove a similar global convergence result for the PINN neural network approximators. Both proofs require analyzing a kernel function in the limit ODE governing the evolution of the limit neural network approximator. A key technical challenge is that the kernel function, which is a composition of the PDE operator and the neural tangent kernel (NTK) operator, lacks a spectral gap, therefore requiring a careful analysis of its properties.

With the advent of large datasets, offline reinforcement learning (RL) is a promising framework for learning good decision-making policies without the need to interact with the real environment. However, offline RL requires the dataset to be reward-annotated, which presents practical challenges when reward engineering is difficult or when obtaining reward annotations is labor-intensive. In this paper, we introduce Optimal Transport Reward labeling (OTR), an algorithm that assigns rewards to offline trajectories, with a few high-quality demonstrations. OTR's key idea is to use optimal transport to compute an optimal alignment between an unlabeled trajectory in the dataset and an expert demonstration to obtain a similarity measure that can be interpreted as a reward, which can then be used by an offline RL algorithm to learn the policy. OTR is easy to implement and computationally efficient. On D4RL benchmarks, we show that OTR with a single demonstration can consistently match the performance of offline RL with ground-truth rewards., Comment: Published in ICLR 2023

Imitation learning (IL) seeks to teach agents specific tasks through expert demonstrations. One of the key approaches to IL is to define a distance between agent and expert and to find an agent policy that minimizes that distance. Optimal transport methods have been widely used in imitation learning as they provide ways to measure meaningful distances between agent and expert trajectories. However, the problem of how to optimally combine multiple expert demonstrations has not been widely studied. The standard method is to simply concatenate state (-action) trajectories, which is problematic when trajectories are multi-modal. We propose an alternative method that uses a multi-marginal optimal transport distance and enables the combination of multiple and diverse state-trajectories in the OT sense, providing a more sensible geometric average of the demonstrations. Our approach enables an agent to learn from several experts, and its efficiency is analyzed on OpenAI Gym control environments and demonstrates that the standard method is not always optimal.

We propose a novel framework for exploring weak and $L_2$ generalization errors of algorithms through the lens of differential calculus on the space of probability measures. Specifically, we consider the KL-regularized empirical risk minimization problem and establish generic conditions under which the generalization error convergence rate, when training on a sample of size $n$, is $\mathcal{O}(1/n)$. In the context of supervised learning with a one-hidden layer neural network in the mean-field regime, these conditions are reflected in suitable integrability and regularity assumptions on the loss and activation functions., Comment: 49 pages

Synthetic data has gained significant momentum thanks to sophisticated machine learning tools that enable the synthesis of high-dimensional datasets. However, many generation techniques do not give the data controller control over what statistical patterns are captured, leading to concerns over privacy protection. While synthetic records are not linked to a particular real-world individual, they can reveal information about users indirectly which may be unacceptable for data owners. There is thus a need to empirically verify the privacy of synthetic data -- a particularly challenging task in high-dimensional data. In this paper we present a general framework for synthetic data generation that gives data controllers full control over which statistical properties the synthetic data ought to preserve, what exact information loss is acceptable, and how to quantify it. The benefits of the approach are that (1) one can generate synthetic data that results in high utility for a given task, while (2) empirically validating that only statistics considered safe by the data curator are used to generate the data. We thus show the potential for synthetic data to be an effective means of releasing confidential data safely, while retaining useful information for analysts.

Personal data collected at scale promises to improve decision-making and accelerate innovation. However, sharing and using such data raises serious privacy concerns. A promising solution is to produce synthetic data, artificial records to share instead of real data. Since synthetic records are not linked to real persons, this intuitively prevents classical re-identification attacks. However, this is insufficient to protect privacy. We here present TAPAS, a toolbox of attacks to evaluate synthetic data privacy under a wide range of scenarios. These attacks include generalizations of prior works and novel attacks. We also introduce a general framework for reasoning about privacy threats to synthetic data and showcase TAPAS on several examples., Comment: Published at the SyntheticData4ML Neurips workshop

Continuous Normalizing Flows (CNFs) are a class of generative models that transform a prior distribution to a model distribution by solving an ordinary differential equation (ODE). We propose to train CNFs on manifolds by minimizing probability path divergence (PPD), a novel family of divergences between the probability density path generated by the CNF and a target probability density path. PPD is formulated using a logarithmic mass conservation formula which is a linear first order partial differential equation relating the log target probabilities and the CNF's defining vector field. PPD has several key benefits over existing methods: it sidesteps the need to solve an ODE per iteration, readily applies to manifold data, scales to high dimensions, and is compatible with a large family of target paths interpolating pure noise and data in finite time. Theoretically, PPD is shown to bound classical probability divergences. Empirically, we show that CNFs learned by minimizing PPD achieve state-of-the-art results in likelihoods and sample quality on existing low-dimensional manifold benchmarks, and is the first example of a generative model to scale to moderately high dimensional manifolds., Comment: ICML 2022

This explainer document aims to provide an overview of the current state of the rapidly expanding work on synthetic data technologies, with a particular focus on privacy. The article is intended for a non-technical audience, though some formal definitions have been given to provide clarity to specialists. This article is intended to enable the reader to quickly become familiar with the notion of synthetic data, as well as understand some of the subtle intricacies that come with it. We do believe that synthetic data is a very useful tool, and our hope is that this report highlights that, while drawing attention to nuances that can easily be overlooked in its deployment., Comment: Commissioned by the Royal Society. 57 pages 2 figures

Solving high-dimensional partial differential equations (PDEs) is a major challenge in scientific computing. We develop a new numerical method for solving elliptic-type PDEs by adapting the Q-learning algorithm in reinforcement learning. Our "Q-PDE" algorithm is mesh-free and therefore has the potential to overcome the curse of dimensionality. Using a neural tangent kernel (NTK) approach, we prove that the neural network approximator for the PDE solution, trained with the Q-PDE algorithm, converges to the trajectory of an infinite-dimensional ordinary differential equation (ODE) as the number of hidden units $\rightarrow \infty$. For monotone PDE (i.e. those given by monotone operators, which may be nonlinear), despite the lack of a spectral gap in the NTK, we then prove that the limit neural network, which satisfies the infinite-dimensional ODE, converges in $L^2$ to the PDE solution as the training time $\rightarrow \infty$. More generally, we can prove that any fixed point of the wide-network limit for the Q-PDE algorithm is a solution of the PDE (not necessarily under the monotone condition). The numerical performance of the Q-PDE algorithm is studied for several elliptic PDEs.

We study the use of amortized optimization to predict optimal transport (OT) maps from the input measures, which we call Meta OT. This helps repeatedly solve similar OT problems between different measures by leveraging the knowledge and information present from past problems to rapidly predict and solve new problems. Otherwise, standard methods ignore the knowledge of the past solutions and suboptimally re-solve each problem from scratch. We instantiate Meta OT models in discrete and continuous settings between grayscale images, spherical data, classification labels, and color palettes and use them to improve the computational time of standard OT solvers. Our source code is available at http://github.com/facebookresearch/meta-ot, Comment: ICML 2023

Samuel A Cohen,1 Suzann Pershing1– 3 1Department of Ophthalmology, Stanford University School of Medicine, Stanford, CA, USA; 2VA Palo Alto Health Care System, Palo Alto, CA, USA; 3Byers Eye Institute at Stanford, Stanford, CA, USACorrespondence: Suzann Pershing, Email pershing@stanford.eduAim: To evaluate and quantify social media presence of ophthalmology peer-reviewed journals, professional societies, and eye health organizations, and to determine if there is a correlation between social media utilization and Twitter engagement metrics or journal impact measures.Methods: We searched for online profiles of 100 ophthalmology peer-reviewed journals, 333 professional ophthalmology societies, and 40 eye health organizations on Facebook, Twitter, and Instagram. Impact was quantified by recording the number of “likes” on Facebook and number of followers on Twitter and Instagram. We also used Twitonomy software to obtain advanced Twitter metrics for all journal accounts from 2018 to 2021, and compared to journal impact measured by SCImago Journal Rank (SJR) score, the h-index, and impact factor.Results: Eye health organizations averaged significantly greater Facebooks “likes” and Twitter followers than both peer-reviewed journals and professional societies (p < 0.0001). Of 100 journals studied, 30% were active on Twitter, 25% on Facebook, and 6% on Instagram. Slightly more than half of all journal-affiliated social media accounts were accessible from the journal website. Among journals with active Twitter accounts, total followers, total tweets, average retweets, and average “favorites” were all significantly positively correlated with academic metrics such as the SJR, h-index, and/or impact factor.Conclusion: Greater social media engagement is associated with higher ophthalmology journal impact metrics; however, ophthalmology journals and professional societies lag behind eye health organizations in social media engagement. Although unable to demonstrat

Prions consist of pathological aggregates of cellular prion protein and have the ability to replicate, causing neurodegenerative diseases, a phenomenon mirrored in many other diseases connected to protein aggregation, including Alzheimer's and Parkinson's diseases. However, despite their key importance in disease, the individual processes governing this formation of pathogenic aggregates, as well as their rates, have remained challenging to elucidate in vivo. Here we bring together a mathematical framework with kinetics of the accumulation of prions in mice and microfluidic measurements of aggregate size to dissect the overall aggregation reaction into its constituent processes and quantify the reaction rates in mice. Taken together, the data show that multiplication of prions in vivo is slower than in in vitro experiments, but efficient when compared with other amyloid systems, and displays scaling behavior characteristic of aggregate fragmentation. These results provide a framework for the determination of the mechanisms of disease-associated aggregation processes within living organisms.

Acetonitrile (CH3CN) is a small, aprotic molecule. The apparent simplicity of this species belies great complexity in its organization and dynamics in the liquid state. Additionally, at silica surfaces, liquid acetonitrile takes on a lipid-bilayer-like structure that has great practical importance in batteries, chromatography, and heterogeneous catalysis. In this thesis, I use molecular simulations and neutron scattering to explore the structure and dynamics of liquid acetonitrile in the bulk and at silica interfaces. First, using angularly resolved radial distribution functions, I identify a microscopic structure in bulk acetonitrile in which most liquid molecules form antiparallel and/or head-to-tail dimers. In contrast to the traditional view of acetonitrile, antiparallel dimers are shown to be octupole-paired, as opposed to dipole-paired. Further analysis reveals that head-to-tail dimers are the dominant motif and live longer than antiparallel dimers. I also use angularly resolved radial distribution functions to show that this propensity to form head-to-dimers has signatures in the observed crystalline polymorphs, and I connect my results using molecular simulation to neutron-scattering data that I also collected. Second, I present findings on the structure and transport properties of acetonitrile at the liquid/silica interface. I use molecular simulations to show that the bilayer is exceptionally robust with large changes in temperature. The effects of Poiseuille hydrodynamic flow on the surface bilayer suggest where the flow boundary may lie, because there is a departure in the vicinity of the walls from the standard parabolic fit describing Poiseuille flow. These results will help guide ion-separation and ion-current experiments using acetonitrile as a solvent. In a separate set of chapters, I present a data analysis protocol for the characterization of absorptive linearities and nonlinearities measured using 2-beam action spectroscopy. This nonlinear-optica

Cross-domain imitation learning studies how to leverage expert demonstrations of one agent to train an imitation agent with a different embodiment or morphology. Comparing trajectories and stationary distributions between the expert and imitation agents is challenging because they live on different systems that may not even have the same dimensionality. We propose Gromov-Wasserstein Imitation Learning (GWIL), a method for cross-domain imitation that uses the Gromov-Wasserstein distance to align and compare states between the different spaces of the agents. Our theory formally characterizes the scenarios where GWIL preserves optimality, revealing its possibilities and limitations. We demonstrate the effectiveness of GWIL in non-trivial continuous control domains ranging from simple rigid transformation of the expert domain to arbitrary transformation of the state-action space., Comment: ICLR 2022

Modeling distributions on Riemannian manifolds is a crucial component in understanding non-Euclidean data that arises, e.g., in physics and geology. The budding approaches in this space are limited by representational and computational tradeoffs. We propose and study a class of flows that uses convex potentials from Riemannian optimal transport. These are universal and can model distributions on any compact Riemannian manifold without requiring domain knowledge of the manifold to be integrated into the architecture. We demonstrate that these flows can model standard distributions on spheres, and tori, on synthetic and geological data. Our source code is freely available online at http://github.com/facebookresearch/rcpm, Comment: ICML 2021

Inverse reinforcement learning attempts to reconstruct the reward function in a Markov decision problem, using observations of agent actions. As already observed in Russell [1998] the problem is ill-posed, and the reward function is not identifiable, even under the presence of perfect information about optimal behavior. We provide a resolution to this non-identifiability for problems with entropy regularization. For a given environment, we fully characterize the reward functions leading to a given policy and demonstrate that, given demonstrations of actions for the same reward under two distinct discount factors, or under sufficiently different environments, the unobserved reward can be recovered up to a constant. We also give general necessary and sufficient conditions for reconstruction of time-homogeneous rewards on finite horizons, and for action-independent rewards, generalizing recent results of Kim et al. [2021] and Fu et al. [2018].

Gaussian processes (GPs) are nonparametric Bayesian models that have been applied to regression and classification problems. One of the approaches to alleviate their cubic training cost is the use of local GP experts trained on subsets of the data. In particular, product-of-expert models combine the predictive distributions of local experts through a tractable product operation. While these expert models allow for massively distributed computation, their predictions typically suffer from erratic behaviour of the mean or uncalibrated uncertainty quantification. By calibrating predictions via a tempered softmax weighting, we provide a solution to these problems for multiple product-of-expert models, including the generalised product of experts and the robust Bayesian committee machine. Furthermore, we leverage the optimal transport literature and propose a new product-of-expert model that combines predictions of local experts by computing their Wasserstein barycenter, which can be applied to both regression and classification., Comment: ICML 2020

Multi-marginal optimal transport enables one to compare multiple probability measures, which increasingly finds application in multi-task learning problems. One practical limitation of multi-marginal transport is computational scalability in the number of measures, samples and dimensionality. In this work, we propose a multi-marginal optimal transport paradigm based on random one-dimensional projections, whose (generalized) distance we term the sliced multi-marginal Wasserstein distance. To construct this distance, we introduce a characterization of the one-dimensional multi-marginal Kantorovich problem and use it to highlight a number of properties of the sliced multi-marginal Wasserstein distance. In particular, we show that (i) the sliced multi-marginal Wasserstein distance is a (generalized) metric that induces the same topology as the standard Wasserstein distance, (ii) it admits a dimension-free sample complexity, (iii) it is tightly connected with the problem of barycentric averaging under the sliced-Wasserstein metric. We conclude by illustrating the sliced multi-marginal Wasserstein on multi-task density estimation and multi-dynamics reinforcement learning problems.

The Asymptotic Randomised Control (ARC) algorithm provides a rigorous approximation to the optimal strategy for a wide class of Bayesian bandits, while retaining low computational complexity. In particular, the ARC approach provides nearly optimal choices even when the payoffs are correlated or more than the reward is observed. The algorithm is guaranteed to asymptotically optimise the expected discounted payoff, with error depending on the initial uncertainty of the bandit. In this paper, we extend the ARC framework to consider a batched bandit problem where observations arrive from a generalised linear model. In particular, we develop a large sample approximation to allow correlated and generally distributed observation. We apply this to a classic dynamic pricing problem based on a Bayesian hierarchical model and demonstrate that the ARC algorithm outperforms alternative approaches.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients, mostly consisting of a variety of essential oils and botanical extracts. This publication, seventh in the series, re-evaluates NFCs with constituent profiles dominated by phenolic derivatives including carvacrol, thymol and related compounds using a constituent-based procedure first published in 2005 and updated in 2018. The procedure is based on the chemical characterization of each NFC as intended for commerce and the estimated intake of the constituent congeneric groups. The procedure applies the threshold of toxicological concern (TTC) concept and evaluates relevant data on absorption, metabolism, genotoxic potential and toxicology of the constituent congeneric groups and the NFC under evaluation. Herein, the FEMA Expert Panel affirmed the generally recognized as safe (GRAS) status of seven phenolic derivative-based NFCs, Origanum Oil (Extractive) (FEMA 2828), Savory Summer Oil (FEMA 3013), Savory Summer Oleoresin (FEMA 3014), Savory Winter Oil (FEMA 3016), Savory Winter Oleoresin (FEMA 3017), Thyme Oil (FEMA 3064) and Thyme White Oil (FEMA 3065) under their conditions of intended use as flavor ingredients.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a program for the re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients. This publication, the sixth in the series, will summarize the re-evaluation of eight NFCs whose constituent profiles are characterized by significant amounts of eucalyptol and/or other cyclic ethers. This re-evaluation was based on a procedure first published in 2005 and subsequently updated in 2018 that evaluates the safety of naturally occurring mixtures for their intended use as flavoring ingredients. The procedure relies on a complete chemical characterization of the NFC intended for commerce and the organization of its chemical constituents into well-defined congeneric groups. The safety of the NFC is evaluated using the well-established and conservative threshold of toxicological concern (TTC) concept in addition to data on absorption, metabolism and toxicology of the constituents of the congeneric groups and the NFC under evaluation. Eight NFCs derived from the Eucalyptus, Melaleuca, Origanum, Laurus, Rosmarinus and Salvia genera were affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association initiated the safety re-evaluation of over 250 natural flavor complexes (NFCs) used as flavor ingredients. This publication, 4th in a series focusing on the safety evaluation of NFCs, presents an evaluation of NFCs rich in hydroxyallylbenzene and hydroxypropenylbenzene constituents using a procedure initially published in 2005 and updated in 2018 that evaluates the safety of naturally occurring mixtures for their intended use as flavoring ingredients. The procedure requires the characterization of the chemical composition for each NFC and subsequent organization of the constituents into defined congeneric groups. The safety of each NFC is evaluated using the conservative threshold of toxicological concern (TTC) approach together with studies on absorption, metabolism and toxicology of the NFC and its constituent congeneric groups. By the application of this procedure, seven NFCs, derived from clove, cinnamon leaf and West Indian bay leaf were affirmed as “generally recognized as safe (GRAS)” under their conditions of intended use as flavor ingredients. An eighth NFC, an oleoresin of West Indian bay leaf, was affirmed based on its estimated intake, which is below the TTC of 0.15 μg/person per day for compounds with structural alerts for genotoxicity.

Barycentric averaging is a principled way of summarizing populations of measures. Existing algorithms for estimating barycenters typically parametrize them as weighted sums of Diracs and optimize their weights and/or locations. However, these approaches do not scale to high-dimensional settings due to the curse of dimensionality. In this paper, we propose a scalable and general algorithm for estimating barycenters of measures in high dimensions. The key idea is to turn the optimization over measures into an optimization over generative models, introducing inductive biases that allow the method to scale while still accurately estimating barycenters. We prove local convergence under mild assumptions on the discrepancy showing that the approach is well-posed. We demonstrate that our method is fast, achieves good performance on low-dimensional problems, and scales to high-dimensional settings. In particular, our approach is the first to be used to estimate barycenters in thousands of dimensions., Comment: In submission

Dynamic time warping (DTW) is a useful method for aligning, comparing and combining time series, but it requires them to live in comparable spaces. In this work, we consider a setting in which time series live on different spaces without a sensible ground metric, causing DTW to become ill-defined. To alleviate this, we propose Gromov dynamic time warping (GDTW), a distance between time series on potentially incomparable spaces that avoids the comparability requirement by instead considering intra-relational geometry. We demonstrate its effectiveness at aligning, combining and comparing time series living on incomparable spaces. We further propose a smoothed version of GDTW as a differentiable loss and assess its properties in a variety of settings, including barycentric averaging, generative modeling and imitation learning.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a program for the re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients. This publication, third in the series, considers NFCs composed primarily of constituents with the 3-phenyl-2-propenyl or a cinnamyl functional group, using the procedure outlined in 2005 and updated in 2018 to evaluate the safety of naturally-occurring mixtures for their intended use as flavor ingredients. The procedure relies on a complete chemical characterization of the NFC intended for commerce and organization of each NFC's chemical constituents into well-defined congeneric groups. The safety of the NFC is evaluated using the well-established and conservative threshold of toxicological concern (TTC) concept in addition to data on absorption, metabolism and toxicology of members of the congeneric groups and the NFC under evaluation. Six NFCs from the Myroxylon and Cinnamomum genera, Balsam Oil, Peru (FEMA 2117), Tolu Balsam Extract (FEMA 3069), Cassia Bark Extract (FEMA 2257), Cassia Bark Oil (FEMA 2258), Cinnamon Bark Extract (FEMA 2290) and Cinnamon Bark Oil (FEMA 2291) were evaluated and affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients. NFC flavor materials include a variety of essential oils and botanical extracts. The re-evaluation of NFCs is conducted based on a constituent-based procedure outlined in 2005 and updated in 2018 that evaluates the safety of NFCs for their intended use as flavor ingredients. This procedure is applied in the re-evaluation of the generally recognized as safe (GRAS) status of NFCs with constituent profiles that are dominated by alicyclic ketones such as menthone and carvone, secondary alcohols such as menthol and carveol, and related compounds. The FEMA Expert Panel affirmed the GRAS status of Peppermint Oil (FEMA 2848), Spearmint Oil (FEMA 3032), Spearmint Extract (FEMA 3031), Cornmint Oil (FEMA 4219), Erospicata Oil (FEMA 4777), Curly Mint Oil (FEMA 4778), Pennyroyal Oil (FEMA 2839), Buchu Leaves Oil (FEMA 2169), Caraway Oil (FEMA 2238) and Dill Oil (FEMA 2383) and determined FEMA GRAS status for Buchu Leaves Extract (FEMA 4923), Peppermint Oil, Terpeneless (FEMA 4924) and Spearmint Oil, Terpeneless (FEMA 4925).

The Flavor and Extract Manufacturers Association (FEMA) Expert Panel relies on the weight of evidence from all available data in the safety evaluation of flavoring substances. This process includes data from genotoxicity studies designed to assess the potential of a chemical agent to react with DNA or otherwise cause changes to DNA, either in vitro or in vivo. The Panel has reviewed a large number of in vitro and in vivo genotoxicity studies during the course of its ongoing safety evaluations of flavorings. The adherence of genotoxicity studies to standardized protocols and guidelines, the biological relevance of the results from those studies, and the human relevance of these studies are all important considerations in assessing whether the results raise specific concerns for genotoxic potential. The Panel evaluates genotoxicity studies not only for evidence of genotoxicity hazard, but also for the probability of risk to the consumer in the context of exposure from their use as flavoring substances. The majority of flavoring substances have given no indication of genotoxic potential in studies evaluated by the FEMA Expert Panel. Examples illustrating the assessment of genotoxicity data for flavoring substances and the consideration of the factors noted above are provided. The weight of evidence approach adopted by the FEMA Expert Panel leads to a rational assessment of risk associated with consumer intake of flavoring substances under the conditions of use.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a program for the re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavor ingredients. This publication, fifth in the series, evaluates the safety of NFCs containing linalool and/or other characteristic mono- and sesquiterpenoid tertiary alcohols and esters using the safety evaluation procedure published by the FEMA Expert Panel in 2005 and updated in 2018. The procedure relies on a complete chemical characterization of the NFC intended for commerce and organization of the chemical constituents of each NFC into well-defined congeneric groups. The safety of each NFC is evaluated using the well-established and conservative threshold of toxicological concern (TTC) concept in addition to data on absorption, metabolism and toxicology of both the constituent congeneric groups and the NFCs. Sixteen NFCs, derived from the Lavandula, Aniba, Elettaria, Daucus, Salvia, Coriandrum, Ribes, Guaiacum/Bulnesia, Citrus, Pogostemon, Melaleuca and Michelia genera, were affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavor ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

We consider stochastic volatility models under parameter uncertainty and investigate how model derived prices of European options are affected. We let the pricing parameters evolve dynamically in time within a specified region, and formalise the problem as a control problem where the control acts on the parameters to maximise/minimise the option value. Through a dual representation with backward stochastic differential equations, we obtain explicit equations for Heston’s model and investigate several numerical solutions thereof. In an empirical study, we apply our results to market data from the S&P 500 index where the model is estimated to historical asset prices. We find that the conservative model-prices cover 98% of the considered market-prices for a set of European call options.

We relate models based on costs of switching beliefs (e.g., due to inattention) to hypothesis tests. Specifically, for an inference problem with a penalty for mistakes and for switching the inferred value, a band of inaction is optimal. We show this band is equivalent to a confidence interval, and therefore to a two-sided hypothesis test.

We consider stochastic volatility models under parameter uncertainty and investigate how model derived prices of European options are affected. We let the pricing parameters evolve dynamically in time within a specified region, and formalise the problem as a control problem where the control acts on the parameters to maximise/minimise the option value. Through a dual representation with backward stochastic differential equations, we obtain explicit equations for Heston’s model and investigate several numerical solutions thereof. In an empirical study, we apply our results to market data from the S&P 500 index where the model is estimated to historical asset prices. We find that the conservative model-prices cover 98% of the considered market-prices for a set of European call options.

In 2015, the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) initiated a re-evaluation of the safety of over 250 natural flavor complexes (NFCs) used as flavoring ingredients. This publication is the first in a series and summarizes the evaluation of 54 Citrus-derived NFCs using the procedure outlined in Smith et al. (2005) and updated in Cohen et al. (2018) to evaluate the safety of naturally-occurring mixtures for their intended use as flavoring ingredients. The procedure relies on a complete chemical characterization of each NFC intended for commerce and organization of each NFC's chemical constituents into well-defined congeneric groups. The safety of the NFC is evaluated using the well-established and conservative threshold of toxicological concern (TTC) concept in addition to data on absorption, metabolism and toxicology of members of the congeneric groups and the NFC under evaluation. As a result of the application of the procedure, 54 natural flavor complexes derived from botanicals of the Citrus genus were affirmed as generally recognized as safe (GRAS) under their conditions of intended use as flavoring ingredients based on an evaluation of each NFC and the constituents and congeneric groups therein.

In the FAME! Project, a code-switching (CS) automatic speech recognition (ASR) system for Frisian-Dutch speech is developed that can accurately transcribe the local broadcaster's bilingual archives with CS speech. This archive contains recordings with monolingual Frisian and Dutch speech segments as well as Frisian-Dutch CS speech, hence the recognition performance on monolingual segments is also vital for accurate transcriptions. In this work, we propose a multi-graph decoding and rescoring strategy using bilingual and monolingual graphs together with a unified acoustic model for CS ASR. The proposed decoding scheme gives the freedom to design and employ alternative search spaces for each (monolingual or bilingual) recognition task and enables the effective use of monolingual resources of the high-resourced mixed language in low-resourced CS scenarios. In our scenario, Dutch is the high-resourced and Frisian is the low-resourced language. We therefore use additional monolingual Dutch text resources to improve the Dutch language model (LM) and compare the performance of single- and multi-graph CS ASR systems on Dutch segments using larger Dutch LMs. The ASR results show that the proposed approach outperforms baseline single-graph CS ASR systems, providing better performance on the monolingual Dutch segments without any accuracy loss on monolingual Frisian and code-mixed segments., Comment: Accepted for publication at Interspeech 2019

The aggregates of the Ab peptide associated with Alzheimer's disease are able to both grow in size aswell as generate, through secondary nucleation, new small oligomeric species, that are major cytotoxins associated with neuronal death. Despite the importance of these amyloid fibril-dependent processes, their structural and molecular underpinnings have remained challenging to elucidate. Here, we consider two molecular chaperones: The Brichos domain, which suppresses specifically secondary nucleation processes, and clusterin which our results show is capable of inhibiting, specifically, the elongation of Ab fibrils at remarkably low substoichiometric ratios. Microfluidic diffusional sizing measurements demonstrate that this inhibition originates from interactions of clusterin with fibril ends with high affinity. Kinetic experiments in the presence of both molecular chaperones reveal that their inhibitory effects are additive and noncooperative, thereby indicating that the reactive sites associated with the formation of new aggregates and the growth of existing aggregates are distinct.

An effective and thorough approach for the safety evaluation of natural flavor complexes (NFCs) was published in 2005 by the Expert Panel of the Flavor and Extract Manufacturers Association (FEMA). An updated procedure is provided here, which maintains the essential concepts of the use of the congeneric group approach and the reliance on the threshold of toxicological concern (TTC) concept. The updated procedure emphasizes more rigorous considerations of unidentified constituents and the genotoxic potential of constituents. The update of the previously established procedure is the first step in a multi-year project to conduct safety re-evaluations for more than 250 NFCs that have uses that are currently considered Generally Recognized as Safe (GRAS) by the FEMA Expert Panel. In addition, this procedure can be more generally employed in the safety evaluation of NFCs.

Background and aimsUse of GI endoscopy is historically lower in nonwhite ethnic and racial groups compared with whites. These disparities are multifactorial but likely contribute to differences in GI clinical outcomes. We sought to improve endoscopy uptake overall and in minorities by predictive overbooking and active recruitment in a hospital-based GI clinic.MethodsFrom January to October 2014, we alternated between traditional booking for Veterans Affairs Healthcare Network patients with a physician recommendation for endoscopy and active recruitment of patients to fill projected open endoscopy appointment slots. On intervention weeks, patients attending a GI clinic were given the opportunity to "fast track" to an endoscopy appointment on short notice. During control weeks, patients were not actively recruited. We compared uptake of endoscopy appointments in both groups and performed logistic regression to determine predictors of participation in fast-track active recruitment.ResultsDuring fast-track active recruitment for endoscopy, the clinic recruited an additional 111 patients for endoscopy over passive recruitment, including 46 African Americans (41.4%). In a logistic regression model controlling for demographic and clinical characteristics, African Americans were twice as likely (adjusted OR, 1.99; 95% CI, 1.26-3.17) than whites to participate in the fast-track option for recommended endoscopy.ConclusionsInterventions to actively recruit patients for endoscopy increased the overall percentage of GI clinic patients undergoing endoscopy and disproportionately improved endoscopy appointment uptake in African Americans.

This publication is the second in a series by the Expert Panel of the Flavor and Extract Manufacturers Association summarizing the conclusions of its third systematic re-evaluation of the safety of flavorings previously considered to be generally recognized as safe (GRAS) under conditions of intended use. Re-evaluation of GRAS status for flavorings is based on updated considerations of exposure, structural analogy, metabolism, pharmacokinetics and toxicology and includes a comprehensive review of the scientific information on the flavorings and structurally related substances. Of the 12 substituted thiophenes reviewed here, 11 were reaffirmed as GRAS based on their rapid absorption, metabolism and excretion in humans and animals; the low estimated dietary exposure from flavor use; the wide margins of safety between the conservative estimates of intake and the no-observed-adverse effect levels; and the lack of significant genotoxic and mutagenic potential. For one of the substituted thiophenes, 3-acetyl-2,5-dimethylthiophene, it was concluded that more detailed exposure information, comparative metabolism studies and comprehensive toxicity data, including an in-depth evaluation of the mechanism of action for any adverse effects observed, are required for continuation of its FEMA GRAS™ status. In the absence of these data, the compound was removed from the FEMA GRAS list.

p-Mentha-1,8-dien-7-al is a naturally occurring cyclic alpha,beta-unsaturated aldehyde that is used as a flavoring substance throughout the world. Due to the chemical structure and the potential DNA reactivity of the alpha,beta-unsaturated carbonyl moiety, a battery of genotoxicity assays was requested by the European Food Safety Authority. Previous genotoxicity studies on the substance gave mixed results, but both positive and negative results were hampered by not always being performed to any standard guideline. The new test battery data indicated some evidence of mutagenicity in vitro, but an in vivo comet/micronucleus combination assay performed in rats was concluded by the study directors to not result in any biologically relevant positive responses. However, EFSA concluded that the in vivo assay gave evidence that p-mentha-1,8-dien-7-al was of potential genotoxic concern. The Expert Panel of the Flavor and Extract Manufacturers Association (FEMA) has reviewed the newly available data and considered its interpretation relative to standard guidelines such as that established by the Organization for Economic Cooperation and Development, and has concluded that the results in the comet/micronucleus combination assay are consistent with the interpretation by the study directors; namely, that p-mentha-1,8-dien-7-al does not appear to have any in vivo genotoxic potential.

Ure2p is the protein determinant of the Saccharomyces cerevisiae prion state [URE3]. Constitutive overexpression of the HSP70 family member SSA1 cures cells of [URE3]. Here, we show that SSA1 increases the lag time of Ure2p fibril formation in vitro in the presence or absence of nucleotide. The presence of the HSP40 co-chaperone Ydj1p has an additive effect on the inhibition of Ure2p fibril formation, whereas the Ydj1p H34Q mutant shows reduced inhibition alone and in combination with Ssa1p. In order to investigate the structural basis of these effects, we constructed and tested an Ssa1p mutant lacking the ATPase domain, as well as a series of C-terminal truncation mutants. The results indicate that Ssa1p can bind to Ure2p and delay fibril formation even in the absence of the ATPase domain, but interaction of Ure2p with the substrate-binding domain is strongly influenced by the C-terminal lid region. Dynamic light scattering, quartz crystal microbalance assays, pull-down assays and kinetic analysis indicate that Ssa1p interacts with both native Ure2p and fibril seeds, and reduces the rate of Ure2p fibril elongation in a concentration-dependent manner. These results provide new insights into the structural and mechanistic basis for inhibition of Ure2p fibril formation by Ssa1p and Ydj1p.