Grounding human-object interaction to affordance behavior in multimodal datasets.

While affordance detection and Human-Object interaction (HOI) detection tasks are related, the theoretical foundation of affordances makes it clear that the two are distinct. In particular, researchers in affordances make distinctions between J. J. Gibson's traditional definition of an affordance, "the action possibilities" of the object within the environment, and the definition of a telic affordance, or one defined by conventionalized purpose or use. We augment the HICO-DET dataset with annotations for Gibsonian and telic affordances and a subset of the dataset with annotations for the orientation of the humans and objects involved. We then train an adapted Human-Object Interaction (HOI) model and evaluate a pre-trained viewpoint estimation system on this augmented dataset. Our model, AffordanceUPT, is based on a two-stage adaptation of the Unary-Pairwise Transformer (UPT), which we modularize to make affordance detection independent of object detection. Our approach exhibits generalization to new objects and actions, can effectively make the Gibsonian/telic distinction, and shows that this distinction is correlated with features in the data that are not captured by the HOI annotations of the HICO-DET dataset.

On the asymptotic behavior of the average geodesic distance L and the compactness CB of simple connected undirected graphs whose order approaches infinity.

The average geodesic distance L Newman (2003) and the compactness CB Botafogo (1992) are important graph indices in applications of complex network theory to real-world problems. Here, for simple connected undirected graphs G of order n, we study the behavior of L(G) and CB(G), subject to the condition that their order |V(G)| approaches infinity. We prove that the limit of L(G)/n and CB(G) lies within the interval [0;1/3] and [2/3;1], respectively. Moreover, for any not necessarily rational number ß ∈ [0;1/3] (α ∈ [2/3;1]) we show how to construct the sequence of graphs {G}, |V(G)| = n → ∞, for which the limit of L(G)/n (CB(G)) is exactly ß (α) (Theorems 1 and 2). Based on these results, our work points to novel classification possibilities of graphs at the node level as well as to the information-theoretic classification of the structural complexity of graph indices.

Second Floor of Flatland: Epitaxial Growth of Graphene on Hexagonal Boron Nitride.

In the studies presented here, the subsequent growth of graphene on hexagonal boron nitride (h-BN) is achieved by the thermal decomposition of molecular precursors and the catalytic assistance of metal substrates. The epitaxial growth of h-BN on Pt(111) is followed by the deposition of a temporary Pt film that acts as a catalyst for the fabrication of the graphene sheet. After intercalation of the intermediate Pt film underneath the boron-nitride mesh, graphene resides on top of h-BN. Scanning tunneling microscopy and density functional calculations reveal that the moiré pattern of the van-der-Waals-coupled double layer is due to the interface of h-BN and Pt(111). While on Pt(111) the graphene honeycomb unit cells uniformly appear as depressions using a clean metal tip for imaging, on h-BN they are arranged in a honeycomb lattice where six protruding unit cells enframe a topographically dark cell. This superstructure is most clearly observed at small probe-surface distances. Spatially resolved inelastic electron tunneling spectroscopy enables the detection of a previously predicted acoustic hybrid phonon of the stacked materials. Its' spectroscopic signature is visible in surface regions where the single graphene sheet on Pt(111) transitions into the top layer of the stacking.

Scanning tunneling microscopy and spectroscopy of rubrene on clean and graphene-covered metal surfaces.

Rubrene (C42H28) was adsorbed with submonolayer coverage on Pt(111), Au(111), and graphene-covered Pt(111). Adsorption phases and vibronic properties of C42H28 consistently reflect the progressive reduction of the molecule-substrate hybridization. Separate C42H28 clusters are observed on Pt(111) as well as broad molecular resonances. On Au(111) and graphene-covered Pt(111) compact molecular islands with similar unit cells of the superstructure characterize the adsorption phase. The highest occupied molecular orbital of C42H28 on Au(111) exhibits weak vibronic progression while unoccupied molecular resonances appear with a broad line shape. In contrast, vibronic subbands are present for both frontier orbitals of C42H28 on graphene. They are due to different molecular vibrational quanta with distinct Huang-Rhys factors.

Dissimilar Decoupling Behavior of Two-Dimensional Materials on Metal Surfaces.

The efficiency of hexagonal boron nitride and graphene to separate the hydrocarbon molecule C64H36 from Ru(0001) and Pt(111) surfaces is explored in low-temperature scanning tunneling microscopy and spectroscopy experiments. Both 2D materials enable the observation of the Franck-Condon effect in both frontier orbitals. On hexagonal boron nitride, vibronic progression with two vibrational energies gives rise to sharp orbital sidebands that are clearly visible up to the second order of the vibrational quantum number with different Huang-Rhys factors. In contrast, on graphene, orbital and vibronic spectroscopic signatures exhibit broad line shapes, with the second-order progression being hardly discriminable. Only a single vibrational quantum energy leaves its fingerprint in the Franck-Condon spectrum.

CRFVoter: gene and protein related object recognition using a conglomerate of CRF-based tools.

BACKGROUND: Gene and protein related objects are an important class of entities in biomedical research, whose identification and extraction from scientific articles is attracting increasing interest. In this work, we describe an approach to the BioCreative V.5 challenge regarding the recognition and classification of gene and protein related objects. For this purpose, we transform the task as posed by BioCreative V.5 into a sequence labeling problem. We present a series of sequence labeling systems that we used and adapted in our experiments for solving this task. Our experiments show how to optimize the hyperparameters of the classifiers involved. To this end, we utilize various algorithms for hyperparameter optimization. Finally, we present CRFVoter, a two-stage application of Conditional Random Field (CRF) that integrates the optimized sequence labelers from our study into one ensemble classifier. RESULTS: We analyze the impact of hyperparameter optimization regarding named entity recognition in biomedical research and show that this optimization results in a performance increase of up to 60%. In our evaluation, our ensemble classifier based on multiple sequence labelers, called CRFVoter, outperforms each individual extractor's performance. For the blinded test set provided by the BioCreative organizers, CRFVoter achieves an F-score of 75%, a recall of 71% and a precision of 80%. For the GPRO type 1 evaluation, CRFVoter achieves an F-Score of 73%, a recall of 70% and achieved the best precision (77%) among all task participants. CONCLUSION: CRFVoter is effective when multiple sequence labeling systems are to be used and performs better then the individual systems collected by it.

LSTMVoter: chemical named entity recognition using a conglomerate of sequence labeling tools.

BACKGROUND: Chemical and biomedical named entity recognition (NER) is an essential preprocessing task in natural language processing. The identification and extraction of named entities from scientific articles is also attracting increasing interest in many scientific disciplines. Locating chemical named entities in the literature is an essential step in chemical text mining pipelines for identifying chemical mentions, their properties, and relations as discussed in the literature. In this work, we describe an approach to the BioCreative V.5 challenge regarding the recognition and classification of chemical named entities. For this purpose, we transform the task of NER into a sequence labeling problem. We present a series of sequence labeling systems that we used, adapted and optimized in our experiments for solving this task. To this end, we experiment with hyperparameter optimization. Finally, we present LSTMVoter, a two-stage application of recurrent neural networks that integrates the optimized sequence labelers from our study into a single ensemble classifier. RESULTS: We introduce LSTMVoter, a bidirectional long short-term memory (LSTM) tagger that utilizes a conditional random field layer in conjunction with attention-based feature modeling. Our approach explores information about features that is modeled by means of an attention mechanism. LSTMVoter outperforms each extractor integrated by it in a series of experiments. On the BioCreative IV chemical compound and drug name recognition (CHEMDNER) corpus, LSTMVoter achieves an F1-score of 90.04%; on the BioCreative V.5 chemical entity mention in patents corpus, it achieves an F1-score of 89.01%. AVAILABILITY AND IMPLEMENTATION: Data and code are available at https://github.com/texttechnologylab/LSTMVoter .

Preparation of graphene bilayers on platinum by sequential chemical vapour deposition.

A cheap and flexible method is introduced that enables the epitaxial growth of bilayer graphene on Pt(111) by sequential chemical vapour deposition. Extended regions of two stacked graphene sheets are obtained by, first, the thermal decomposition of ethylene and the subsequent formation of graphene. In the second step, a sufficiently thick Pt film buries the first graphene layer and acts as a platform for the fabrication of the second graphene layer in the third step. A final annealing process then leads to the diffusion of the first graphene sheet to the surface until the bilayer stacking with the second sheet is accomplished. Scanning tunnelling microscopy unravels the successful growth of bilayer graphene and elucidates the origin of moiré patterns.

On the limit value of compactness of some graph classes.

In this paper, we study the limit of compactness which is a graph index originally introduced for measuring structural characteristics of hypermedia. Applying compactness to large scale small-world graphs (Mehler, 2008) observed its limit behaviour to be equal 1. The striking question concerning this finding was whether this limit behaviour resulted from the specifics of small-world graphs or was simply an artefact. In this paper, we determine the necessary and sufficient conditions for any sequence of connected graphs resulting in a limit value of CB = 1 which can be generalized with some consideration for the case of disconnected graph classes (Theorem 3). This result can be applied to many well-known classes of connected graphs. Here, we illustrate it by considering four examples. In fact, our proof-theoretical approach allows for quickly obtaining the limit value of compactness for many graph classes sparing computational costs.

Assessing cognitive alignment in different types of dialog by means of a network model.

We present a network model of dialog lexica, called TiTAN (Two-layer Time-Aligned Network) series. TiTAN series capture the formation and structure of dialog lexica in terms of serialized graph representations. The dynamic update of TiTAN series is driven by the dialog-inherent timing of turn-taking. The model provides a link between neural, connectionist underpinnings of dialog lexica on the one hand and observable symbolic behavior on the other. On the neural side, priming and spreading activation are modeled in terms of TiTAN networking. On the symbolic side, TiTAN series account for cognitive alignment in terms of the structural coupling of the linguistic representations of dialog partners. This structural stance allows us to apply TiTAN in machine learning of data of dialogical alignment. In previous studies, it has been shown that aligned dialogs can be distinguished from non-aligned ones by means of TiTAN -based modeling. Now, we simultaneously apply this model to two types of dialog: task-oriented, experimentally controlled dialogs on the one hand and more spontaneous, direction giving dialogs on the other. We ask whether it is possible to separate aligned dialogs from non-aligned ones in a type-crossing way. Starting from a recent experiment (Mehler, Lücking, & Menke, 2011a), we show that such a type-crossing classification is indeed possible. This hints at a structural fingerprint left by alignment in networks of linguistic items that are routinely co-activated during conversation.