Search

The detection, identification, and localization of illicit radiological and nuclear material continue to be key components of nuclear non-proliferation and nuclear security efforts around the world. Networks of radiation detectors deployed at strategic locations in urban environments have the potential to provide continuous radiological/nuclear (R/N) surveillance and provide high probabilities of intercepting threat sources. The integration of contextual information from sensors such as video, Lidar, and meteorological sensors can provide significantly enhanced situational awareness, and improved detection and localization performance through the fusion of the radiological and contextual data. In this work, we present details of our work to establish a city-scale multi-sensor network testbed for intelligent, adaptive R/N detection in urban environments, and develop new techniques that enable city-scale source detection, localization, and tracking.

The ability to anticipate technical expertise and capability evolution trends globally is essential for national and global security, especially in safety-critical domains like nuclear nonproliferation (NN) and rapidly emerging fields like artificial intelligence (AI). In this work, we extend traditional statistical relational learning approaches (e.g., link prediction in collaboration networks) and formulate a problem of anticipating technical expertise and capability evolution using dynamic heterogeneous graph representations. We develop novel capabilities to forecast collaboration patterns, authorship behavior, and technical capability evolution at different granularities (e.g., scientist and institution levels) in two distinct research fields. We implement a dynamic graph transformer (DGT) neural architecture, which pushes the state-of-the-art graph neural network models by (a) forecasting heterogeneous (rather than homogeneous) nodes and edges, and (b) relying on both discrete -- and continuous -- time inputs. We demonstrate that our DGT models predict collaboration, partnership, and expertise patterns with 0.26, 0.73, and 0.53 mean reciprocal rank values for AI and 0.48, 0.93, and 0.22 for NN domains. DGT model performance exceeds the best-performing static graph baseline models by 30-80% across AI and NN domains. Our findings demonstrate that DGT models boost inductive task performance, when previously unseen nodes appear in the test data, for the domains with emerging collaboration patterns (e.g., AI). Specifically, models accurately predict which established scientists will collaborate with early career scientists and vice-versa in the AI domain.

Dynamic Mode Decomposition (DMD) describes complex dynamic processes through a hierarchy of simpler coherent features. DMD is regularly used to understand the fundamental characteristics of turbulence and is closely related to Koopman operators. However, verifying the decomposition, equivalently the computed spectral features of Koopman operators, remains a major challenge due to the infinite-dimensional nature of Koopman operators. Challenges include spurious (unphysical) modes, and dealing with continuous spectra, both of which occur regularly in turbulent flows. Residual Dynamic Mode Decomposition (ResDMD), introduced by (Colbrook & Townsend 2021), overcomes some of these challenges through the data-driven computation of residuals associated with the full infinite-dimensional Koopman operator. ResDMD computes spectra and pseudospectra of general Koopman operators with error control, and computes smoothed approximations of spectral measures (including continuous spectra) with explicit high-order convergence theorems. ResDMD thus provides robust and verified Koopmanism. We implement ResDMD and demonstrate its application in a variety of fluid dynamic situations, at varying Reynolds numbers, arising from both numerical and experimental data. Examples include: vortex shedding behind a cylinder; hot-wire data acquired in a turbulent boundary layer; particle image velocimetry data focusing on a wall-jet flow; and acoustic pressure signals of laser-induced plasma. We present some advantages of ResDMD, namely, the ability to verifiably resolve non-linear, transient modes, and spectral calculation with reduced broadening effects. We also discuss how a new modal ordering based on residuals enables greater accuracy with a smaller dictionary than the traditional modulus ordering. This paves the way for greater dynamic compression of large datasets without sacrificing accuracy.

Machine learning models that learn from dynamic graphs face nontrivial challenges in learning and inference as both nodes and edges change over time. The existing large-scale graph benchmark datasets that are widely used by the community primarily focus on homogeneous node and edge attributes and are static. In this work, we present a variety of large scale, dynamic heterogeneous academic graphs to test the effectiveness of models developed for multi-step graph forecasting tasks. Our novel datasets cover both context and content information extracted from scientific publications across two communities: Artificial Intelligence (AI) and Nuclear Nonproliferation (NN). In addition, we propose a systematic approach to improve the existing evaluation procedures used in the graph forecasting models.

We develop a rapid and accurate contour method for the solution of time-fractional PDEs. The method inverts the Laplace transform via an optimised stable quadrature rule, suitable for infinite-dimensional operators, whose error decreases like $\exp(-cN/\log(N))$ for $N$ quadrature points. The method is parallisable, avoids having to resolve singularities of the solution as $t\downarrow 0$, and avoids the large memory consumption that can be a challenge for time-stepping methods applied to time-fractional PDEs. The ODEs resulting from quadrature are solved using adaptive sparse spectral methods that converge exponentially with optimal linear complexity. These solutions of ODEs are reused for different times. We provide a complete analysis of our approach for fractional beam equations used to model small-amplitude vibration of viscoelastic materials with a fractional Kelvin-Voigt stress-strain relationship. We calculate the system's energy evolution over time and the surface deformation in cases of both constant and non-constant viscoelastic parameters. An infinite-dimensional ``solve-then-discretise'' approach considerably simplifies the analysis, which studies the generalisation of the numerical range of a quasi-linearisation of a suitable operator pencil. This allows us to build an efficient algorithm with explicit error control. The approach can be readily adapted to other time-fractional PDEs and is not constrained to fractional parameters in the range $0<\nu<1$.

Width-based planning has shown promising results on Atari 2600 games using pixel input, while using substantially fewer environment interactions than reinforcement learning. Recent width-based approaches have computed feature vectors for each screen using a hand designed feature set or a variational autoencoder trained on game screens (VAE-IW), and prune screens that do not have novel features during the search. We propose Olive (Online-VAE-IW), which updates the VAE features online using active learning to maximize the utility of screens observed during planning. Experimental results in 55 Atari games demonstrate that it outperforms Rollout-IW by 42-to-11 and VAE-IW by 32-to-20. Moreover, Olive outperforms existing work based on policy-learning ($\pi$-IW, DQN) trained with 100x training budget by 30-to-22 and 31-to-17, and a state of the art data-efficient reinforcement learning (EfficientZero) trained with the same training budget and ran with 1.8x planning budget by 18-to-7 in Atari 100k benchmark, with no policy learning at all. The source code is available at github.com/ibm/atari-active-learning ., Comment: Accepted in ICAPS 2022, Planning & Learning track. Edits: Title change. Fixed the variance update formula. Updated the scores of $\pi$-IW as per authors' request. Included the results of DQN and EfficientZero

Objective: To investigate whether performance (number of correct decisions) of humans supported by a computer alerting tool can be improved by tailoring the tool's alerting threshold (sensitivity/specificity combination) according to user ability and task difficulty. Background: Many researchers implicitly assume that for each tool there exists a single ideal threshold. But research shows the effects of alerting tools on decision errors to vary depending on variables such as user ability and task difficulty. These findings motivated our investigation. Method: Forty-seven participants edited text passages, aided by a simulated spell-checker tool. We experimentally manipulated passage difficulty and tool alerting threshold, measured participants' editing and dictation ability, and counted participants' decision errors (false positives + false negatives). We investigated whether alerting threshold, user ability, task difficulty and their interactions affected error count. Results: Which alerting threshold better helped a user depended on an interaction between user ability and passage difficulty. Some effects were large: for higher ability users, a more sensitive threshold reduced errors by 30%, on the easier passages. Participants were not significantly more likely to prefer the alerting threshold with which they performed better. Conclusion: Adjusting alerting thresholds for individual users' ability and task difficulty could substantially improve effects of automated alerts on user decisions. This potential deserves further investigation. Improvement size and rules for adjustment will be application-specific. Application: Computer alerting tools have critical roles in many domains. Designers should assess potential benefits of adjustable alerting thresholds for their specific CAT application. Guidance for choosing thresholds will be essential for realizing these benefits in practice., Comment: 25 pages, 5 figures

Deceptive news posts shared in online communities can be detected with NLP models, and much recent research has focused on the development of such models. In this work, we use characteristics of online communities and authors -- the context of how and where content is posted -- to explain the performance of a neural network deception detection model and identify sub-populations who are disproportionately affected by model accuracy or failure. We examine who is posting the content, and where the content is posted to. We find that while author characteristics are better predictors of deceptive content than community characteristics, both characteristics are strongly correlated with model performance. Traditional performance metrics such as F1 score may fail to capture poor model performance on isolated sub-populations such as specific authors, and as such, more nuanced evaluation of deception detection models is critical.

Evaluating model robustness is critical when developing trustworthy models not only to gain deeper understanding of model behavior, strengths, and weaknesses, but also to develop future models that are generalizable and robust across expected environments a model may encounter in deployment. In this paper we present a framework for measuring model robustness for an important but difficult text classification task - deceptive news detection. We evaluate model robustness to out-of-domain data, modality-specific features, and languages other than English. Our investigation focuses on three type of models: LSTM models trained on multiple datasets(Cross-Domain), several fusion LSTM models trained with images and text and evaluated with three state-of-the-art embeddings, BERT ELMo, and GloVe (Cross-Modality), and character-level CNN models trained on multiple languages (Cross-Language). Our analyses reveal a significant drop in performance when testing neural models on out-of-domain data and non-English languages that may be mitigated using diverse training data. We find that with additional image content as input, ELMo embeddings yield significantly fewer errors compared to BERT orGLoVe. Most importantly, this work not only carefully analyzes deception model robustness but also provides a framework of these analyses that can be applied to new models or extended datasets in the future.

With the increasing use of machine-learning driven algorithmic judgements, it is critical to develop models that are robust to evolving or manipulated inputs. We propose an extensive analysis of model robustness against linguistic variation in the setting of deceptive news detection, an important task in the context of misinformation spread online. We consider two prediction tasks and compare three state-of-the-art embeddings to highlight consistent trends in model performance, high confidence misclassifications, and high impact failures. By measuring the effectiveness of adversarial defense strategies and evaluating model susceptibility to adversarial attacks using character- and word-perturbed text, we find that character or mixed ensemble models are the most effective defenses and that character perturbation-based attack tactics are more successful.

Evaluation beyond aggregate performance metrics, e.g. F1-score, is crucial to both establish an appropriate level of trust in machine learning models and identify future model improvements. In this paper we demonstrate CrossCheck, an interactive visualization tool for rapid crossmodel comparison and reproducible error analysis. We describe the tool and discuss design and implementation details. We then present three use cases (named entity recognition, reading comprehension, and clickbait detection) that show the benefits of using the tool for model evaluation. CrossCheck allows data scientists to make informed decisions to choose between multiple models, identify when the models are correct and for which examples, investigate whether the models are making the same mistakes as humans, evaluate models' generalizability and highlight models' limitations, strengths and weaknesses. Furthermore, CrossCheck is implemented as a Jupyter widget, which allows rapid and convenient integration into data scientists' model development workflows.

We present a constructive procedure for the calculation of 2-D potential flows in periodic domains with multiple boundaries per period window. The solution requires two steps: (i) a conformal mapping from a canonical circular domain to the physical target domain, and (ii) the construction of the complex potential inside the circular domain. All singly periodic domains may be classified into three distinct types: unbounded in two directions, unbounded in one direction, and bounded. In each case, we relate the target periodic domain to a canonical circular domain via conformal mapping and present the functional form of prototypical conformal maps for each type of target domain. We then present solutions for a range of potential flow phenomena including flow singularities, moving boundaries, uniform flows, straining flows and circulatory flows. By phrasing the solutions in terms of the transcendental Schottky--Klein prime function, the ensuing solutions are valid for an arbitrary number of obstacles per period window. Moreover, our solutions are exact and do not require any asymptotic approximations., Comment: 29 pages; 7 figures; submitted to TCFD in Dec 2019

"Ground effect" refers to the enhanced performance enjoyed by fliers or swimmers operating close to the ground. We derive a number of exact solutions for this phenomenon, thereby elucidating the underlying physical mechanisms involved in ground effect. Unlike previous analytic studies, our solutions are not restricted to particular parameter regimes such as "weak" or "extreme" ground effect, and do not even require thin aerofoil theory. Moreover, the solutions are valid for a hitherto intractable range of flow phenomena including point vortices, uniform and straining flows, unsteady motions of the wing, and the Kutta condition. We model the ground effect as the potential flow past a wing inclined above a flat wall. The solution of the model requires two steps: firstly, a coordinate transformation between the physical domain and a concentric annulus, and secondly, the solution of the potential flow problem inside the annulus. We show that both steps can be solved by introducing a new special function which is straightforward to compute. Moreover, the ensuing solutions are simple to express and offer new insight into the mathematical structure of ground effect. In order to identify the missing physics in our potential flow model, we compare our solutions against new experimental data. The experiments show that boundary layer separation on the wing and wall occurs at small angles of attack, and we suggest ways in which our model could be extended to account for these effects., Comment: Main body: 10 pages & 3 figures; supplementary material: 6 pages & 5 figures. Submitted to JFM Rapids

We present a solution for the scattered field caused by an incident wave interacting with an infinite cascade of blades with complex boundary conditions. This extends previous studies by allowing the blades to be compliant, porous or satisfying a generalised impedance condition. Beginning with the convected wave equation, we employ Fourier transforms to obtain an integral equation amenable to the Wiener--Hopf method. This Wiener--Hopf system is solved using a method that avoids the factorisation of matrix functions. The Fourier transform is inverted to obtain an expression for the acoustic potential function that is valid throughout the entire domain. We observe that the principal effect of complex boundary conditions is to perturb the zeros of the Wiener--Hopf kernel, which correspond to the duct modes in the inter-blade region. We focus efforts on understanding the role of porosity, and present a range of results on sound transmission and generation. The behaviour of the duct modes is discussed in detail in order to explain the physical mechanisms behind the associated noise reductions. In particular, we show that cut-on duct modes do not exist for arbitrary porosity coefficients. Conversely, the acoustic modes are unchanged by modifications to the boundary conditions. Consequently, we observe that even modest values of porosity can result in reductions in the sound power level of $5$ dB for the first mode and $20$ dB for the second mode. The solution is essentially analytic (the only numerical requirements are matrix inversion and root finding) and is therefore extremely rapid to compute., Comment: 40 pages

Our main contribution in this work is novel results of multilingual models that go beyond typical applications of rumor or misinformation detection in English social news content to identify fine-grained classes of digital deception across multiple languages (e.g. Russian, Spanish, etc.). In addition, we present models for multimodal deception detection from images and text and discuss the limitations of image only and text only models. Finally, we elaborate on the ongoing work on measuring deceptive content (in particular disinformation) spread across social platforms.

Leading- and trailing-edge serrations have been widely used to reduce the leading- and trailing-edge noise in applications such as contra-rotating fans and large wind turbines. Recent studies show that these two noise problems can be modelled analytically using the Wiener-Hopf method. However, the resulting models involve infinite-interval integrals that cannot be evaluated analytically, and consequently implementing them poses practical difficulty. This paper develops easily-implementable noise prediction models for flat plates with serrated leading and trailing edges, respectively. By exploiting the fact that high-order modes are cut-off and adjacent modes do not interfere in the far field except at sufficiently high frequencies, an infinite-interval integral involving two infinite sums is approximated by a single straightforward sum. Numerical comparison shows that the resulting models serve as excellent approximations to the original models. Good agreement is also achieved when the leading-edge model predictions are compared with experimental results for sawtooth serrations of various root-to-tip amplitudes. Importantly, the models developed in this paper can be evaluated robustly in a very efficient manner. For example, a typical far-field noise spectrum can be calculated within milliseconds for both the trailing- and leading-edge noise models on a standard desktop computer. Due to their efficiency and ease of numerical implementation, these models are expected to be of particular importance in applications where a numerical optimization is likely to be needed., Comment: 24 pages, 10 figures