Showing papers by "University of Massachusetts Amherst published in 2017"

Memristors with diffusive dynamics as synaptic emulators for neuromorphic computing

Abstract: The accumulation and extrusion of Ca2+ in the pre- and postsynaptic compartments play a critical role in initiating plastic changes in biological synapses. To emulate this fundamental process in electronic devices, we developed diffusive Ag-in-oxide memristors with a temporal response during and after stimulation similar to that of the synaptic Ca2+ dynamics. In situ high-resolution transmission electron microscopy and nanoparticle dynamics simulations both demonstrate that Ag atoms disperse under electrical bias and regroup spontaneously under zero bias because of interfacial energy minimization, closely resembling synaptic influx and extrusion of Ca2+, respectively. The diffusive memristor and its dynamics enable a direct emulation of both short- and long-term plasticity of biological synapses, representing an advance in hardware implementation of neuromorphic functionalities.

Association analysis identifies 65 new breast cancer risk loci

Abstract: Breast cancer risk is influenced by rare coding variants in susceptibility genes, such as BRCA1, and many common, mostly non-coding variants. However, much of the genetic contribution to breast cancer risk remains unknown. Here we report the results of a genome-wide association study of breast cancer in 122,977 cases and 105,974 controls of European ancestry and 14,068 cases and 13,104 controls of East Asian ancestry. We identified 65 new loci that are associated with overall breast cancer risk at P < 5 × 10-8. The majority of credible risk single-nucleotide polymorphisms in these loci fall in distal regulatory elements, and by integrating in silico data to predict target genes in breast cells at each locus, we demonstrate a strong overlap between candidate target genes and somatic driver genes in breast tumours. We also find that heritability of breast cancer due to all single-nucleotide polymorphisms in regulatory features was 2-5-fold enriched relative to the genome-wide average, with strong enrichment for particular transcription factor binding sites. These results provide further insight into genetic susceptibility to breast cancer and will improve the use of genetic risk scores for individualized screening and prevention.

Diverse Applications of Nanomedicine

Abstract: The design and use of materials in the nanoscale size range for addressing medical and health-related issues continues to receive increasing interest. Research in nanomedicine spans a multitude of areas, including drug delivery, vaccine development, antibacterial, diagnosis and imaging tools, wearable devices, implants, high-throughput screening platforms, etc. using biological, nonbiological, biomimetic, or hybrid materials. Many of these developments are starting to be translated into viable clinical products. Here, we provide an overview of recent developments in nanomedicine and highlight the current challenges and upcoming opportunities for the field and translation to the clinic.

Sampling Matters in Deep Embedding Learning

Abstract: Deep embeddings answer one simple question: How similar are two images? Learning these embeddings is the bedrock of verification, zero-shot learning, and visual search. The most prominent approaches optimize a deep convolutional network with a suitable loss function, such as contrastive loss or triplet loss. While a rich line of work focuses solely on the loss functions, we show in this paper that selecting training examples plays an equally important role. We propose distance weighted sampling, which selects more informative and stable examples than traditional approaches. In addition, we show that a simple margin based loss is sufficient to outperform all other loss functions. We evaluate our approach on the Stanford Online Products, CAR196, and the CUB200-2011 datasets for image retrieval and clustering, and on the LFW dataset for face verification. Our method achieves state-of-the-art performance on all of them.

Exploring the Sensitivity of Next Generation Gravitational Wave Detectors

Abstract: The second-generation of gravitational-wave detectors are just starting operation, and have already yielding their first detections. Research is now concentrated on how to maximize the scientific potential of gravitational-wave astronomy. To support this effort, we present here design targets for a new generation of detectors, which will be capable of observing compact binary sources with high signal-to-noise ratio throughout the Universe.

Propensity Score Matching in Accounting Research

Abstract: Propensity score matching (PSM) has become a popular technique for estimating average treatment effects (ATEs) in accounting research. In this study, we discuss the usefulness and limitations of PSM relative to more traditional multiple regression (MR) analysis. We discuss several PSM design choices and review the use of PSM in 86 articles in leading accounting journals from 2008–2014. We document a significant increase in the use of PSM from zero studies in 2008 to 26 studies in 2014. However, studies often oversell the capabilities of PSM, fail to disclose important design choices, and/or implement PSM in a theoretically inconsistent manner. We then empirically illustrate complications associated with PSM in three accounting research settings. We first demonstrate that when the treatment is not binary, PSM tends to confine analyses to a subsample of observations where the effect size is likely to be smallest. We also show that seemingly innocuous design choices greatly influence sample composi...

Meta Networks

Abstract: Neural networks have been successfully applied in applications with a large amount of labeled data However, the task of rapid generalization on new concepts with small training data while preserving performances on previously learned ones still presents a significant challenge to neural network models In this work, we introduce a novel meta learning method, Meta Networks (MetaNet), that learns a meta-level knowledge across tasks and shifts its inductive biases via fast parameterization for rapid generalization When evaluated on Omniglot and Mini-ImageNet benchmarks, our MetaNet models achieve a near human-level performance and outperform the baseline approaches by up to 6% accuracy We demonstrate several appealing properties of MetaNet relating to generalization and continual learning

Biodiversity in the city: key challenges for urban green space management

Abstract: Cities play important roles in the conservation of global biodiversity, particularly through the planning and management of urban green spaces (UGS). However, UGS management is subject to a complex assortment of interacting social, cultural, and economic factors, including governance, economics, social networks, multiple stakeholders, individual preferences, and social constraints. To help deliver more effective conservation outcomes in cities, we identify major challenges to managing biodiversity in UGS and important topics warranting further investigation. Biodiversity within UGS must be managed at multiple scales while accounting for various socioeconomic and cultural influences. Although the environmental consequences of management activities to enhance urban biodiversity are now beginning to be addressed, additional research and practical management strategies must be developed to balance human needs and perceptions while maintaining ecological processes.

Human-started wildfires expand the fire niche across the United States

Abstract: The economic and ecological costs of wildfire in the United States have risen substantially in recent decades. Although climate change has likely enabled a portion of the increase in wildfire activity, the direct role of people in increasing wildfire activity has been largely overlooked. We evaluate over 1.5 million government records of wildfires that had to be extinguished or managed by state or federal agencies from 1992 to 2012, and examined geographic and seasonal extents of human-ignited wildfires relative to lightning-ignited wildfires. Humans have vastly expanded the spatial and seasonal "fire niche" in the coterminous United States, accounting for 84% of all wildfires and 44% of total area burned. During the 21-y time period, the human-caused fire season was three times longer than the lightning-caused fire season and added an average of 40,000 wildfires per year across the United States. Human-started wildfires disproportionally occurred where fuel moisture was higher than lightning-started fires, thereby helping expand the geographic and seasonal niche of wildfire. Human-started wildfires were dominant (>80% of ignitions) in over 5.1 million km2, the vast majority of the United States, whereas lightning-started fires were dominant in only 0.7 million km2, primarily in sparsely populated areas of the mountainous western United States. Ignitions caused by human activities are a substantial driver of overall fire risk to ecosystems and economies. Actions to raise awareness and increase management in regions prone to human-started wildfires should be a focus of United States policy to reduce fire risk and associated hazards.

Face Detection with the Faster R-CNN

Abstract: While deep learning based methods for generic object detection have improved rapidly in the last two years, most approaches to face detection are still based on the R-CNN framework [11], leading to limited accuracy and processing speed. In this paper, we investigate applying the Faster RCNN [26], which has recently demonstrated impressive results on various object detection benchmarks, to face detection. By training a Faster R-CNN model on the large scale WIDER face dataset [34], we report state-of-the-art results on the WIDER test set as well as two other widely used face detection benchmarks, FDDB and the recently released IJB-A.

Performance of the ATLAS trigger system in 2015

Abstract: During 2015 the ATLAS experiment recorded 3.8 fb(-1) of proton-proton collision data at a centre-of-mass energy of 13 TeV. The ATLAS trigger system is a crucial component of the experiment, respons ...

Long-term pattern and magnitude of soil carbon feedback to the climate system in a warming world

Abstract: In a 26-year soil warming experiment in a mid-latitude hardwood forest, we documented changes in soil carbon cycling to investigate the potential consequences for the climate system. We found that soil warming results in a four-phase pattern of soil organic matter decay and carbon dioxide fluxes to the atmosphere, with phases of substantial soil carbon loss alternating with phases of no detectable loss. Several factors combine to affect the timing, magnitude, and thermal acclimation of soil carbon loss. These include depletion of microbially accessible carbon pools, reductions in microbial biomass, a shift in microbial carbon use efficiency, and changes in microbial community composition. Our results support projections of a long-term, self-reinforcing carbon feedback from mid-latitude forests to the climate system as the world warms.

A Deep Relevance Matching Model for Ad-hoc Retrieval

Abstract: In recent years, deep neural networks have led to exciting breakthroughs in speech recognition, computer vision, and natural language processing (NLP) tasks. However, there have been few positive results of deep models on ad-hoc retrieval tasks. This is partially due to the fact that many important characteristics of the ad-hoc retrieval task have not been well addressed in deep models yet. Typically, the ad-hoc retrieval task is formalized as a matching problem between two pieces of text in existing work using deep models, and treated equivalent to many NLP tasks such as paraphrase identification, question answering and automatic conversation. However, we argue that the ad-hoc retrieval task is mainly about relevance matching while most NLP matching tasks concern semantic matching, and there are some fundamental differences between these two matching tasks. Successful relevance matching requires proper handling of the exact matching signals, query term importance, and diverse matching requirements. In this paper, we propose a novel deep relevance matching model (DRMM) for ad-hoc retrieval. Specifically, our model employs a joint deep architecture at the query term level for relevance matching. By using matching histogram mapping, a feed forward matching network, and a term gating network, we can effectively deal with the three relevance matching factors mentioned above. Experimental results on two representative benchmark collections show that our model can significantly outperform some well-known retrieval models as well as state-of-the-art deep matching models.

US Cosmic Visions: New Ideas in Dark Matter 2017: Community Report

Abstract: This white paper summarizes the workshop "U.S. Cosmic Visions: New Ideas in Dark Matter" held at University of Maryland on March 23-25, 2017.

Syntrophy Goes Electric: Direct Interspecies Electron Transfer

Abstract: Direct interspecies electron transfer (DIET) has biogeochemical significance, and practical applications that rely on DIET or DIET-based aspects of microbial physiology are growing. Mechanisms for DIET have primarily been studied in defined cocultures in which Geobacter species are one of the DIET partners. Electrically conductive pili (e-pili) can be an important electrical conduit for DIET. However, there may be instances in which electrical contacts are made between electron transport proteins associated with the outer membranes of the partners. Alternatively, DIET partners can plug into conductive carbon materials, such as granular activated carbon, carbon cloth, and biochar, for long-range electron exchange without the need for e-pili. Magnetite promotes DIET, possibly by acting as a substitute for outer-surface c-type cytochromes. DIET is the primary mode of interspecies electron exchange in some anaerobic digesters converting wastes to methane. Promoting DIET with conductive materials shows promise for stabilizing and accelerating methane production in digesters, permitting higher organic loading rates. Various lines of evidence suggest that DIET is important in terrestrial wetlands, which are an important source of atmospheric methane. DIET may also have a role in anaerobic methane oxidation coupled to sulfate reduction, an important control on methane releases. The finding that DIET can serve as the source of electrons for anaerobic photosynthesis further broadens its potential environmental significance. Microorganisms capable of DIET are good catalysts for several bioelectrochemical technologies and e-pili are a promising renewable source of electronic materials. The study of DIET is in its early stages, and additional investigation is required to better understand the diversity of microorganisms that are capable of DIET, the importance of DIET to carbon and electron flow in anaerobic environments, and the biochemistry and physiology of DIET.

Topological cell clustering in the ATLAS calorimeters and its performance in LHC Run 1

Abstract: The reconstruction of the signal from hadrons and jets emerging from the proton–proton collisions at the Large Hadron Collider (LHC) and entering the ATLAS calorimeters is based on a three-dimensional topological clustering of individual calorimeter cell signals. The cluster formation follows cell signal-significance patterns generated by electromagnetic and hadronic showers. In this, the clustering algorithm implicitly performs a topological noise suppression by removing cells with insignificant signals which are not in close proximity to cells with significant signals. The resulting topological cell clusters have shape and location information, which is exploited to apply a local energy calibration and corrections depending on the nature of the cluster. Topological cell clustering is established as a well-performing calorimeter signal definition for jet and missing transverse momentum reconstruction in ATLAS.

Indeterminate Pronouns: The View from Japanese

Abstract: The quantificational system in Japanese makes use of so-called indeterminate pronouns, which take on existential, universal, interrogative, negative polarity, or free choice interpretations depending on what operator they associate with. Similar systems are found crosslinguistically, which raises the question as to what makes such system look so different from more familiar determiner quantification systems. This paper takes a first step toward answering this question by presenting an analysis of the German indeterminate pronoun or determiner irgendein from a Japanese point of view.

Black Carbon (Biochar) In Water/Soil Environments: Molecular Structure, Sorption, Stability, and Potential Risk.

Abstract: Black carbon (BC) is ubiquitous in the environments and participates in various biogeochemical processes. Both positive and negative effects of BC (especially biochar) on the ecosystem have been identified, which are mainly derived from its diverse physicochemical properties. Nevertheless, few studies systematically examined the linkage between the evolution of BC molecular structure with the resulted BC properties, environmental functions as well as potential risk, which is critical for understanding the BC environmental behavior and utilization as a multifunctional product. Thus, this review highlights the molecular structure evolution of BC during pyrolysis and the impact of BC physicochemical properties on its sorption behavior, stability, and potential risk in terrestrial and aqueous ecosystems. Given the wide application of BC and its important role in biogeochemical processes, future research should focus on the following: (1) establishing methodology to more precisely predict and design BC propert...

Fast and Accurate Entity Recognition with Iterated Dilated Convolutions

Abstract: Today when many practitioners run basic NLP on the entire web and large-volume traffic, faster methods are paramount to saving time and energy costs. Recent advances in GPU hardware have led to the emergence of bi-directional LSTMs as a standard method for obtaining per-token vector representations serving as input to labeling tasks such as NER (often followed by prediction in a linear-chain CRF). Though expressive and accurate, these models fail to fully exploit GPU parallelism, limiting their computational efficiency. This paper proposes a faster alternative to Bi-LSTMs for NER: Iterated Dilated Convolutional Neural Networks (ID-CNNs), which have better capacity than traditional CNNs for large context and structured prediction. Unlike LSTMs whose sequential processing on sentences of length N requires O(N) time even in the face of parallelism, ID-CNNs permit fixed-depth convolutions to run in parallel across entire documents. We describe a distinct combination of network structure, parameter sharing and training procedures that enable dramatic 14-20x test-time speedups while retaining accuracy comparable to the Bi-LSTM-CRF. Moreover, ID-CNNs trained to aggregate context from the entire document are more accurate than Bi-LSTM-CRFs while attaining 8x faster test time speeds.

Direct Cytosolic Delivery of CRISPR/Cas9-Ribonucleoprotein for Efficient Gene Editing

Abstract: Genome editing through the delivery of CRISPR/Cas9-ribonucleoprotein (Cas9-RNP) reduces unwanted gene targeting and avoids integrational mutagenesis that can occur through gene delivery strategies. Direct and efficient delivery of Cas9-RNP into the cytosol followed by translocation to the nucleus remains a challenge. Here, we report a remarkably highly efficient (∼90%) direct cytoplasmic/nuclear delivery of Cas9 protein complexed with a guide RNA (sgRNA) through the coengineering of Cas9 protein and carrier nanoparticles. This construct provides effective (∼30%) gene editing efficiency and opens up opportunities in studying genome dynamics.

Ultra-selective high-flux membranes from directly synthesized zeolite nanosheets

Abstract: A direct synthesis of high-aspect-ratio microporous zeolite nanosheets and the use of such nanosheets in separation membranes are described. Zeolites—naturally occurring porous crystalline aluminosilicates that are also produced industrially on a large scale—are used commercially as selective adsorbents. Zeolite membranes can be used for selective dehydration, but more general separations, for example of hydrocarbon isomers, are challenging because they require thin membranes with highly oriented pores. At present, such thin membranes are produced by an expensive and low-yielding exfoliation process. Here the authors produce nanometre-thick zeolite nanosheets using a bottom-up seeded growth method that retains the pore structure and avoids rotational intergrowths. Using xylene isomer separation as a benchmark, the authors found that their compact membranes had higher selectivities and flux rates than previous zeolite membranes. A zeolite with structure type MFI1,2 is an aluminosilicate or silicate material that has a three-dimensionally connected pore network, which enables molecular recognition in the size range 0.5–0.6 nm. These micropore dimensions are relevant for many valuable chemical intermediates, and therefore MFI-type zeolites are widely used in the chemical industry as selective catalysts or adsorbents3,4,5. As with all zeolites, strategies to tailor them for specific applications include controlling their crystal size and shape5,6,7,8. Nanometre-thick MFI crystals (nanosheets) have been introduced in pillared9 and self-pillared (intergrown)10 architectures, offering improved mass-transfer characteristics for certain adsorption and catalysis applications11,12,13,14. Moreover, single (non-intergrown and non-layered) nanosheets have been used to prepare thin membranes15,16 that could be used to improve the energy efficiency of separation processes17. However, until now, single MFI nanosheets have been prepared using a multi-step approach based on the exfoliation of layered MFI9,15, followed by centrifugation to remove non-exfoliated particles18. This top-down method is time-consuming, costly and low-yield and it produces fragmented nanosheets with submicrometre lateral dimensions. Alternatively, direct (bottom-up) synthesis could produce high-aspect-ratio zeolite nanosheets, with improved yield and at lower cost. Here we use a nanocrystal-seeded growth method triggered by a single rotational intergrowth to synthesize high-aspect-ratio MFI nanosheets with a thickness of 5 nanometres (2.5 unit cells). These high-aspect-ratio nanosheets allow the fabrication of thin and defect-free coatings that effectively cover porous substrates. These coatings can be intergrown to produce high-flux and ultra-selective MFI membranes that compare favourably with other MFI membranes prepared from existing MFI materials (such as exfoliated nanosheets or nanocrystals).

A deep ALMA image of the Hubble Ultra Deep Field

Abstract: European Research Council via the award of an Advanced Grant; EC [312725]; European Research Council via the award of a Consolidator Grant; UK Science and Technology Facilities Council; FWO Pegasus Marie Curie Fellowship; European Research Council through the Advanced Grant [321302, 669253, 670193]; JSPS KAKENHI [JP15K17604]; Chulalongkorn University's CUniverse (CUAASC); Royal Society

Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learning

Abstract: Knowledge bases (KB), both automatically and manually constructed, are often incomplete --- many valid facts can be inferred from the KB by synthesizing existing information. A popular approach to KB completion is to infer new relations by combinatory reasoning over the information found along other paths connecting a pair of entities. Given the enormous size of KBs and the exponential number of paths, previous path-based models have considered only the problem of predicting a missing relation given two entities or evaluating the truth of a proposed triple. Additionally, these methods have traditionally used random paths between fixed entity pairs or more recently learned to pick paths between them. We propose a new algorithm MINERVA, which addresses the much more difficult and practical task of answering questions where the relation is known, but only one entity. Since random walks are impractical in a setting with combinatorially many destinations from a start node, we present a neural reinforcement learning approach which learns how to navigate the graph conditioned on the input query to find predictive paths. Empirically, this approach obtains state-of-the-art results on several datasets, significantly outperforming prior methods.

Physical and Chemical Stability of Curcumin in Aqueous Solutions and Emulsions: Impact of pH, Temperature, and Molecular Environment

Abstract: The utilization of curcumin as a nutraceutical in food and supplement products is often limited because of its low water solubility, poor chemical stability, and low oral bioavailability. This study examined the impact of pH, storage temperature, and molecular environment on the physical and chemical stability of pure curcumin in aqueous solutions and in oil-in-water emulsions. Unlike naturally occurring curcuminoid mixtures (that contain curcumin, demethoxy-curcumin, and bisdemethoxy-curcumin), pure curcumin was highly unstable to chemical degradation in alkaline aqueous solutions (pH ≥7.0) and tended to crystallize out of aqueous acidic solutions (pH <7). These effects were attributed to changes in the molecular structure of curcumin under different pH conditions. The curcumin crystals formed were relatively large (10–50 μm), which made them prone to rapid sedimentation. The incorporation of curcumin into oil-in-water emulsions (30% MCT, 1 mg curcumin/g MCT, d32 ≈ 298 nm) improved its water dispersibili...

3D Shape Segmentation with Projective Convolutional Networks

Abstract: This paper introduces a deep architecture for segmenting 3D objects into their labeled semantic parts. Our architecture combines image-based Fully Convolutional Networks (FCNs) and surface-based Conditional Random Fields (CRFs) to yield coherent segmentations of 3D shapes. The image-based FCNs are used for efficient view-based reasoning about 3D object parts. Through a special projection layer, FCN outputs are effectively aggregated across multiple views and scales, then are projected onto the 3D object surfaces. Finally, a surface-based CRF combines the projected outputs with geometric consistency cues to yield coherent segmentations. The whole architecture (multi-view FCNs and CRF) is trained end-to-end. Our approach significantly outperforms the existing state-of-the-art methods in the currently largest segmentation benchmark (ShapeNet). Finally, we demonstrate promising segmentation results on noisy 3D shapes acquired from consumer-grade depth cameras.