Publications

Scholars and commentators have debated whether lower-threshold forms of political engagement on social media should be treated as being conducive to higher-threshold modes of political participation or a diversion from them. Drawing on an original survey of a representative sample of Italians who discussed the 2013 election on Twitter, we demonstrate that the more respondents acquire political information via social media and express themselves politically on these platforms, the more they are likely to contact politicians via e-mail, campaign for parties and candidates using social media, and attend offline events to which they were invited online. These results suggest that lower-threshold forms of political engagement on social media do not distract from higher-threshold activities, but are strongly associated with them.

...These events — and the corresponding responses on social media — illustrate what has become increasingly evident: it is almost impossible to think of a major political protest or upheaval occurring without social media being part of both the incident and the ensuing narrative. The Euromaidan protests, which culminated in the flight of President Yanukovych from Ukraine in late February 2014, are a case in point. Indeed, the Ukrainian Euromaidan protest movement may go down in history as the first truly successful social media uprising. Earlier movements labeled social media revolutions subsequently have been criticized for not having had much important activity on social media (Moldova, Arab Spring) or for having had a large social media presence but ultimately failing to make much of a long-term impact as a protest movement (Spain’s Los Indignados, Occupy Wall Street, Gezi Park in Turkey). In Ukraine, a government fell, a region was annexed, a civilian plane was shot down, and what some are calling a civil war continues to this day in the eastern part of the country. Clearly, the movement was consequential and, as we will show, social media usage was widespread and significant.

Although undulatory swimming is observed in many organisms, the neuromuscular basis for undulatory movement patterns is not well understood. To better understand the basis for the generation of these movement patterns, we studied muscle activity in the nematode Caenorhabditis elegans. Caenorhabditis elegans exhibits a range of locomotion patterns: in low viscosity fluids the undulation has a wavelength longer than the body and propagates rapidly, while in high viscosity fluids or on agar media the undulatory waves are shorter and slower. Theoretical treatment of observed behaviour has suggested a large change in force–posture relationships at different viscosities, but analysis of bend propagation suggests that short-range proprioceptive feedback is used to control and generate body bends. How muscles could be activated in a way consistent with both these results is unclear. We therefore combined automated worm tracking with calcium imaging to determine muscle activation strategy in a variety of external substrates. Remarkably, we observed that across locomotion patterns spanning a threefold change in wavelength, peak muscle activation occurs approximately 45° (1/8th of a cycle) ahead of peak midline curvature. Although the location of peak force is predicted to vary widely, the activation pattern is consistent with required force in a model incorporating putative length- and velocity-dependence of muscle strength. Furthermore, a linear combination of local curvature and velocity can match the pattern of activation. This suggests that proprioception can enable the worm to swim effectively while working within the limitations of muscle biomechanics and neural control.

Neural network models of early sensory processing typically reduce the dimensionality of streaming input data. Such networks learn the principal subspace, in the sense of principal component analysis, by adjusting synaptic weights according to activity-dependent learning rules. When derived from a principled cost function, these rules are nonlocal and hence biologically implausible. At the same time, biologically plausible local rules have been postulated rather than derived from a principled cost function. Here, to bridge this gap, we derive a biologically plausible network for subspace learning on streaming data by minimizing a principled cost function. In a departure from previous work, where cost was quantified by the representation, or reconstruction, error, we adopt a multidimensional scaling cost function for streaming data. The resulting algorithm relies only on biologically plausible Hebbian and anti-Hebbian local learning rules. In a stochastic setting, synaptic weights converge to a stationary state, which projects the input data onto the principal subspace. If the data are generated by a nonstationary distribution, the network can track the principal subspace. Thus, our result makes a step toward an algorithmic theory of neural computation.

Although undulatory swimming is observed in many organisms, the neuromuscular basis for undulatory movement patterns is not well understood. To better understand the basis for the generation of these movement patterns, we studied muscle activity in the nematode Caenorhabditis elegans. Caenorhabditis elegans exhibits a range of locomotion patterns: in low viscosity fluids the undulation has a wavelength longer than the body and propagates rapidly, while in high viscosity fluids or on agar media the undulatory waves are shorter and slower. Theoretical treatment of observed behaviour has suggested a large change in force-posture relationships at different viscosities, but analysis of bend propagation suggests that short-range proprioceptive feedback is used to control and generate body bends. How muscles could be activated in a way consistent with both these results is unclear. We therefore combined automated worm tracking with calcium imaging to determine muscle activation strategy in a variety of external substrates. Remarkably, we observed that across locomotion patterns spanning a threefold change in wavelength, peak muscle activation occurs approximately 45° (1/8th of a cycle) ahead of peak midline curvature. Although the location of peak force is predicted to vary widely, the activation pattern is consistent with required force in a model incorporating putative length- and velocity-dependence of muscle strength. Furthermore, a linear combination of local curvature and velocity can match the pattern of activation. This suggests that proprioception can enable the worm to swim effectively while working within the limitations of muscle biomechanics and neural control.

A consistent muscle activation strategy underlies crawling and swimming in Caenorhabditis elegans (PDF Download Available). Available from: https://www.researchgate.net/publication/270342595_A_consistent_muscle_activation_strategy_underlies_crawling_and_swimming_in_Caenorhabditis_elegans [accessed Jul 26, 2017].

Elucidation of transcriptional regulatory networks (TRNs) is a fundamental goal in biology, and one of the most important components of TRNs are transcription factors (TFs), proteins that specifically bind to gene promoter and enhancer regions to alter target gene expression patterns. Advances in genomic technologies as well as advances in computational biology have led to multiple large regulatory network models (directed networks) each with a large corpus of supporting data and gene-annotation. There are multiple possible biological motivations for exploring large regulatory network models, including: validating TF-target gene relationships, figuring out co-regulation patterns, and exploring the coordination of cell processes in response to changes in cell state or environment. Here we focus on queries aimed at validating regulatory network models, and on coordinating visualization of primary data and directed weighted gene regulatory networks. The large size of both the network models and the primary data can make such coordinated queries cumbersome with existing tools and, in particular, inhibits the sharing of results between collaborators. In this work, we develop and demonstrate a web-based framework for coordinating visualization and exploration of expression data (RNA-seq, microarray), network models and gene-binding data (ChIP-seq). Using specialized data structures and multiple coordinated views, we design an efficient querying model to support interactive analysis of the data. Finally, we show the effectiveness of our framework through case studies for the mouse immune system (a dataset focused on a subset of key cellular functions) and a model bacteria (a small genome with high data-completeness).

We present a fast direct solver for the simulation of electromagnetic scattering from an arbitrarily-shaped, large, empty cavity embedded in an infinite perfectly conducting half space. The governing Maxwell equations are reformulated as a well-conditioned second kind integral equation and the resulting linear system is solved in nearly linear time using a hierarchical matrix factorization technique. We illustrate the performance of the scheme with several numerical examples for complex cavity shapes over a wide range of frequencies.

Olshausen and Field (OF) proposed that neural computations in the primary visual cortex (V1) can be partially modelled by sparse dictionary learning. By minimizing the regularized representation error they derived an online algorithm, which learns Gabor-filter receptive fields from a natural image ensemble in agreement with physiological experiments. Whereas the OF algorithm can be mapped onto the dynamics and synaptic plasticity in a single-layer neural network, the derived learning rule is nonlocal - the synaptic weight update depends on the activity of neurons other than just pre- and postsynaptic ones - and hence biologically implausible. Here, to overcome this problem, we derive sparse dictionary learning from a novel cost-function - a regularized error of the symmetric factorization of the input's similarity matrix. Our algorithm maps onto a neural network of the same architecture as OF but using only biologically plausible local learning rules. When trained on natural images our network learns Gabor-filter receptive fields and reproduces the correlation among synaptic weights hard-wired in the OF network. Therefore, online symmetric matrix factorization may serve as an algorithmic theory of neural computation.

Social media use among elites offers a useful avenue for analyzing regime response to protest, especially in countries with some degree of freedom of speech. We examine the frequency and content of Twitter usage among Venezuelan elites in the context of the 2014 protests. This analysis demonstrates that the regime sent more signals during protests but that the content of these messages addressed more topics than it did for opposition elites, especially following acts of regime suppression of opposition-sponsored protests. This observation supports theoretical predictions that noisy public information can make coordination more difficult. Regime elites produced noisy Twitter as part of an explicit strategy to prevent citizens from coordinating on revolution. This analysis of social media use by both pro- and anti-regime elites contributes to the debate over whether social media prevents or promotes regime change.