Small Number Of Examples Research Articles

Robust multi-model adaptation regression with local feature space representation

Learning visual models of object categories generally requires a large number of training examples. We show in this paper that it is possible to uncover much discriminative information about a visual category from a small number of examples with just fewlabeled data. The key insight is that, rather than learning from scratch with the original feature representation, we learn an optimal target model with the local feature space representations by leveraging some prior models pre-trained on other related datasets, no matter how different they might be. This target model can be obtained by learning a robust regression scheme with a multi-source adaptation regularization term, which is formally formulated as a Robust Multi-model Adaptation Regression (RMAR) framework with local feature space representations of the target data. This framework is composed of three core components: (1) local consistency Laplacian regularization by constructing a self-adaptive affinity graph based on local feature space representation (LFSR), which aims to augment the discriminative space of the target domain; (2) robust regression learning scheme with feature selection based on l2,1-norm minimization by explicitly considering the regression noise/outlier as well as the regression residual; and (3) scatter constrained multi-model adaptation regularization for generalizing the established robust regression framework to exploit the prior models, under the assumption that multiple auxiliary discriminative models can help the semi-supervised learning from few labeled samples. Lastly, we further propose an effective extension to RMAR, i.e., ensemble LFSR-graph Laplacians regularization framework. The optimization algorithms for RMAR and its extension can be efficiently solved by the alternating iterative strategy, and the iteration convergence can be theoretically guaranteed. Experiments over several real datasets show the promising performance of our methods compared with several representative state-of-the-art works.

Unstable modes in projection-based reduced-order models: How many can there be, and what do they tell you?

Projection methods provide an appealing way to construct reduced-order models of large-scale linear dynamical systems: they are intuitively motivated and fairly easy to compute. Unfortunately, the resulting reduced models need not inherit the stability of the original system. How many unstable modes can these reduced models have? This note investigates this question, using theory originally motivated by iterative methods for linear algebraic systems and eigenvalue problems, and illustrating the theory with a number of small examples. From these results follow rigorous upper bounds on the number of unstable modes in reduced models generated via orthogonal projection, for both continuous- and discrete-time systems. Can anything be learned from the unstable modes in reduced-order models? Several examples illustrate how such instability can helpfully signal transient growth in the original system.

Norm-Based Binary Search Trees for Speeding Up KNN Big Data Classification

Due to their large sizes and/or dimensions, the classification of Big Data is a challenging task using traditional machine learning, particularly if it is carried out using the well-known K-nearest neighbors classifier (KNN) classifier, which is a slow and lazy classifier by its nature. In this paper, we propose a new approach to Big Data classification using the KNN classifier, which is based on inserting the training examples into a binary search tree to be used later for speeding up the searching process for test examples. For this purpose, we used two methods to sort the training examples. The first calculates the minimum/maximum scaled norm and rounds it to 0 or 1 for each example. Examples with 0-norms are sorted in the left-child of a node, and those with 1-norms are sorted in the right child of the same node; this process continues recursively until we obtain one example or a small number of examples with the same norm in a leaf node. The second proposed method inserts each example into the binary search tree based on its similarity to the examples of the minimum and maximum Euclidean norms. The experimental results of classifying several machine learning big datasets show that both methods are much faster than most of the state-of-the-art methods compared, with competing accuracy rates obtained by the second method, which shows great potential for further enhancements of both methods to be used in practice.

Sample Efficient Learning of Path Following and Obstacle Avoidance Behavior for Quadrotors

In this paper we propose an algorithm for the training of neural network control policies for quadrotors. The learned control policy computes control commands directly from sensor inputs and is hence computationally efficient. An imitation learning algorithm produces a policy that reproduces the behavior of a path following control algorithm with collision avoidance. Due to the generalization ability of neural networks, the resulting policy performs local collision avoidance of unseen obstacles while following a global reference path. The algorithm uses a time-free model predictive path-following controller as a supervisor. The controller generates demonstrations by following few example paths. This enables an easy to implement learning algorithm that is robust to errors of the model used in the model predictive controller. The policy is trained on the real quadrotor, which requires collision-free exploration around the example path. An adapted version of the supervisor is used to enable exploration. Thus, the policy can be trained from a relatively small number of examples on the real quadrotor, making the training sample efficient.

Seneca Collapse for the World’s Human Population?

Most scenarios for the world’s human population predict continued growth into the 22nd century, while some indicate that it could stabilize or begin to fall before 2100. Almost always, decline is seen as not being faster than the preceding growth. Different scenarios are obtained if we consider the human population as a complex system, subject to the general rules that govern complex systems, in particular their tendency to show rapid changes which – in the case of populations – may take the shape of true collapses (defined here as “Seneca Collapses”). The present survey examines a small number of examples of rapid population collapses in the human and in the animal domains. While not pretending to be exhaustive, the data presented here show that biological populations do show rapid “Seneca-style” collapses. So, it is possible that the same phenomenon could occur for the world’s human population.

F85. Deep learning for detection of epileptiform discharges from scalp EEG recordings

Clinical use of computer assisted detection of epileptiform discharges is limited, and visual assessment generally outperforms current computer algorithms. Deep learning is a promising novel approach for the detection of epileptiform discharges, being able to process huge amounts of data and to learn by example. Here, we show pilot results of detecting interictal epileptiform discharges using different deep neural networks. We selected 50 routine 19-channel scalp EEGs from patients with focal epilepsies from our digital database. All epileptiform discharges (n = 1815) were annotaded by an experienced clinical neurophysiologist. Additionally, we added 50 EEGs that were classified as normal. Subsequently, we extracted non-overlapping 2 s epochs from the 100 EEGs, and divided them into a training (n = 41,381 epochs) and test (n = 8775 epochs) set. For validation we used 7 EEGs (47,122 epochs) with 538 focal epileptiform discharges and 12 normal EEGs (n = 11,782 epochs). EEGs (bipolar montage) were filtered between 0.5 and 35 Hz as the only pre-processing step. We implemented several combinations of archetypes and variations of Convolutional and Recurrent Neural Networks. The main structures were composed either of CNNs (4–9 layers) or of CNNs on top of a LSTM layer (50–100 units), where both models were usually complemented by Fully Connected layers (1–3). The use of Dropout (0.2–0.5) always improved the results. We trained all networks in python using Keras on top of Theano, using a CUDA-enabled NVIDIA GPU. The input of the networks were the band-pass filtered 19-channel 2 s epochs. The output of the networks was a probability ([0–1]) that this particular epoch contained an epileptiform discharge; we set the threshold at p = 0.5. Training of the networks took approximately 60 min. The best networks obtained receiver operating characteristics with areas under the curve between 0.86 and 0.92. Validation with the independent set of EEGs, classifying non-overlapping 2 s epochs, allowed detection of epileptiform discharges with sensitivities from 20% to 80% with specificities between 95% and 100% (mean false prediction rate ±0.5/min). In the control EEGs, the number of false positives was 31 out of 11,782 epochs (0.25%, false prediction rate ±4/h). Classification for a routine 30 min EEG, providing annotations for each epoch of 2 s, took less than a few seconds. In this pilot study, we show that deep nets can accurately detect epileptiform discharges with high specificity, after being trained with a relatively small number of examples. The difference between the false prediction rates of control EEGs and EEGs with epileptiform discharges (4/h and 0.5/min, respectively) may in part result from novel features in EEGs from patients with epilepsy, unnoticed by visual assessment, but detected by the network. We foresee that deep nets may outperform humans both in classification accuracy and speed, leading to a fundamental shift in clinical EEG analysis in the next decade.

Disjunctive Program Synthesis: A Robust Approach to Programming by Example

Programming by example (PBE) systems allow end users to easily create programs by providing a few input-output examples to specify their intended task. The system attempts to generate a program in a domain specific language (DSL) that satisfies the given examples. However, a key challenge faced by existing PBE techniques is to ensure the robustness of the programs that are synthesized from a small number of examples, as these programs often fail when applied to new inputs. This is because there can be many possible programs satisfying a small number of examples, and the PBE system has to somehow rank between these candidates and choose the correct one without any further information from the user. In this work we present a different approach to PBE in which the system avoids making a ranking decision at the synthesis stage, by instead synthesizing a disjunctive program that includes the many top-ranked programs as possible alternatives and selects between these different choices upon execution on a new input. This delayed choice brings the important benefit of comparing the possible outputs produced by the different disjuncts on a given input at execution time. We present a generic framework for synthesizing such disjunctive programs in arbitrary DSLs, and describe two concrete implementations of disjunctive synthesis in the practical domains of data extraction from plain text and HTML documents. We present an evaluation showing the significant increase in robustness achieved with our disjunctive approach, as illustrated by an increase from 59% to 93% of tasks for which correct programs can be learnt from a single example.

Learning efficient logic programs

When machine learning programs from data, we ideally want to learn efficient rather than inefficient programs. However, existing inductive logic programming (ILP) techniques cannot distinguish between the efficiencies of programs, such as permutation sort (n!) and merge sort O(n;log;n). To address this limitation, we introduce Metaopt, an ILP system which iteratively learns lower cost logic programs, each time further restricting the hypothesis space. We prove that given sufficiently large numbers of examples, Metaopt converges on minimal cost programs, and our experiments show that in practice only small numbers of examples are needed. To learn minimal time-complexity programs, including non-deterministic programs, we introduce a cost function called tree cost which measures the size of the SLD-tree searched when a program is given a goal. Our experiments on programming puzzles, robot strategies, and real-world string transformation problems show that Metaopt learns minimal cost programs. To our knowledge, Metaopt is the first machine learning approach that, given sufficient numbers of training examples, is guaranteed to learn minimal cost logic programs, including minimal time-complexity programs.

ПЕРСПЕКТИВНЫЕ НЕЙРОСЕТЕВЫЕ АЛГОРИТМЫ РАСПОЗНАВАНИЯ ДИНАМИЧЕСКИХ БИОМЕТРИЧЕСКИХ ОБРАЗОВ В ПРОСТРАНСТВЕ ВЗАИМОЗАВИСИМЫХ ПРИЗНАКОВ

A model of neurons for biometric authentication, capable of efficient processing of highly dependent features, based on the agreement criteria (Gini, Сramer-von-Mises, Kolmogorov-Smirnov, the maximum of intersection areas of probability densities) is proposed. An experiment was performed on comparing the efficiency of neurons based on the proposed model and neurons on the basis of difference and hyperbolic Bayesian functionals capable of processing highly dependent biometric data. Variants of construction of hybrid neural networks, that can be trained on a small number of examples of a biometric pattern (about 20), are suggested. An experiment was conducted to collect dynamic biometric patterns, in the experiment 90 people entered handwritten and voice patterns during a month. Intermediate results on recognition of subjects based on hybrid neural networks were obtained. Number of errors in verification of a signature (handwritten password) was less than 2%, verification of a speaker by a fixed passphrase was less than 6%. The testing was carried out on biometric samples, obtained after some time period after the formation of training sample.

Типы связи вербальных и визуальных компонентов в «городском письме» Иркутска

The question of peculiarities of the language environment of the city is considered. The definition of the term urban message , which the author of the article refers to banners, billboard advertising and signs. Trends in the study of creolized texts in modern linguistics are described. The main reasons for the frequent use of paralinguistic elements in the advertising text in general and in the “urban message” in particular are noted. The links (connectivity) between verbal component and the visual part in the texts of the urban environment of Irkutsk are examined. The relevance of the research is determined by the need to assess the effectiveness of the impact of texts on the audience. The novelty of the work is seen in the fact that for the first time “urban message” as a kind of advertising text acts as a full-fledged object of linguistic research. The results of the Irkutsk “urban message” analysis are presented. Classification of types of semantic relations of verbal and visual components is given: (1) direct correlation, (2) peripheral relevance, (3) contamination (complementarity) of verbal-visual unit and ergonym. It is established that the authors of creolized texts prefer the first type of relation between image and slogan. It is suggested that a small number of examples with other types of links is due to the insufficient level of speech culture of the authors of the “urban message.” Isolated cases of unwarranted (and wrong) use of the element of non-verbal part are identified.

Fast Blended Transformations for Partial Shape Registration

Automatic estimation of skinning transformations is a popular way to deform a single reference shape into a new pose by providing a small number of control parameters. We generalize this approach by efficiently enabling the use of multiple exemplar shapes. Using a small set of representative natural poses, we propose to express an unseen appearance by a low-dimensional linear subspace, specified by a redundant dictionary of weighted vertex positions. Minimizing a nonlinear functional that regulates the example manifold, the suggested approach supports local-rigid deformations of articulated objects, as well as nearly isometric embeddings of smooth shapes. A real-time nonrigid deformation system is demonstrated, and a shape completion and partial registration framework is introduced. These applications can recover a target pose and implicit inverse kinematics from a small number of examples and just a few vertex positions. The resulting reconstruction is more accurate compared to alternative reduced deformable models.

Kinetic resolution of sterically hindered secondary alcohols catalyzed by aminophosphinite organocatalyst

Kinetic resolution of secondary alcohols by benzoylation using a phosphinite derivative of (1S,2R)-1-amino-2-indanol as the catalyst was investigated. The aminophosphinite catalyst is effective for the kinetic resolution of aryl cycloalkyl carbinols with a small number of examples for organocatalytic kinetic resolution to achieve resolution with s = up to 44. Although the benzoylation of phenylalkanols proceeded with a low selectivity, 1-arylalkanols bearing at least one substituent at the ortho position on the benzene ring or a branched alkyl group on the carbinol carbon were resolved with acceptable selectivity.

Enhanced interlaminar excitation or reduced superficial layer inhibition in neocortex generates different spike-and-wave-like electrographic events in vitro.

Acute in vitro models have revealed a great deal of information about mechanisms underlying many types of epileptiform activity. However, few examples exist that shed light on spike-and-wave (SpW) patterns of pathological activity. SpW are seen in many epilepsy syndromes, both generalized and focal, and manifest across the entire age spectrum. They are heterogeneous in terms of their severity, symptom burden, and apparent anatomical origin (thalamic, neocortical, or both), but any relationship between this heterogeneity and underlying pathology remains elusive. In this study we demonstrate that physiological delta-frequency rhythms act as an effective substrate to permit modeling of SpW of cortical origin and may help to address this issue. For a starting point of delta activity, multiple subtypes of SpW could be modeled computationally and experimentally by either enhancing the magnitude of excitatory synaptic events ascending from neocortical layer 5 to layers 2/3 or selectively modifying superficial layer GABAergic inhibition. The former generated SpW containing multiple field spikes with long interspike intervals, whereas the latter generated SpW with short-interval multiple field spikes. Both types had different laminar origins and each disrupted interlaminar cortical dynamics in a different manner. A small number of examples of human recordings from patients with different diagnoses revealed SpW subtypes with the same temporal signatures, suggesting that detailed quantification of the pattern of spikes in SpW discharges may be a useful indicator of disparate underlying epileptogenic pathologies.NEW & NOTEWORTHY Spike-and-wave-type discharges (SpW) are a common feature in many epilepsies. Their electrographic manifestation is highly varied, as are available genetic clues to associated underlying pathology. Using computational and in vitro models, we demonstrate that distinct subtypes of SpW are generated by lamina-selective disinhibition or enhanced interlaminar excitation. These subtypes could be detected in at least some noninvasive patient recordings, suggesting more detailed analysis of SpW may be useful in determining clinical pathology.

Postsynthetic Tuning of Metal-Organic Frameworks for Targeted Applications.

Metal-organic frameworks (MOFs) are periodic, hybrid, atomically well-defined porous materials that typically form by self-assembly and consist of inorganic nodes (metal ions or clusters) and multitopic organic linkers. MOFs as a whole offer many intriguing properties, including ultrahigh porosity, tunable chemical functionality, and low density. These properties point to numerous potential applications, including gas storage, chemical separations, catalysis, light harvesting, and chemical sensing, to name a few. Reticular chemistry, or the linking of molecular building blocks into predetermined network structures, has been employed to synthesize thousands of MOFs. Given the vast library of candidate nodes and linkers, the number of potentially synthetically accessible MOFs is enormous. Nevertheless, a powerful complementary approach to obtain specific structures with desired chemical functionality is to modify known MOFs after synthesis. This approach is particularly useful when incorporation of particular chemical functionalities via direct synthesis is challenging or impossible. The challenges may stem from limited stability or solubility of precursors, unwanted secondary reactivity of precursors, or incompatibility of functional groups with the conditions needed for direct synthesis. MOFs can be postsynthetically modified by replacing the metal nodes and/or organic linkers or via functionalization of the metal nodes and/or organic linkers. Here we describe some of our efforts toward the development and application of postsynthetic strategies for imparting desired chemical functionalities in MOFs of known topology. The techniques include methods for functionalizing MOF nodes, i.e., solvent-assisted ligand incorporation (SALI) and atomic layer deposition in MOFs (AIM) as well as a method to replace structural linkers, termed solvent-assisted linker exchange (SALE), also known as postsynthethic exchange (PSE). For each functionalization strategy, we first describe its chemical basis along with the requirements for its successful implementation. We then present a small number of examples, with an emphasis on those that (a) convey the underlying concepts and/or (b) lead to functional structures (e.g., catalysts) that would be difficult or impossible to access via direct routes. The examples, however, are only illustrative, and a significant body of work exists from both our lab and others, especially for the SALE/PSE strategy. We refer readers to the papers cited and to the references therein. More exciting, in our view, will be new examples and new applications of the functionalization strategies-especially applications made possible by creatively combining the strategies. Underexplored (again, in our view) are implementations that impart electrical conductivity, enable increasingly selective chemical sensing, or facilitate cascade catalysis. It will be interesting to see where these strategies and others take this compelling field over the next few years.

Inverse electron energy dispersion from moving auroral forms

AbstractNumerous published examples of energy‐dispersed bursts show electron energies reaching as high as several keV and decaying to lower energies over a fraction of 1 s. This signature has been interpreted by some authors as due to impulsive acceleration to a broad range of energies in a localized region and by others as the result of impulsive, dispersive Alfvén waves, in which case the acceleration takes place over an extended distance along magnetic field lines. A survey by the Suprathermal (0–350 eV) Electron Imager on the Enhanced Polar Outflow Probe (ePOP) in the topside ionosphere has produced examples of high‐to‐low (“regular”) energy dispersion, but also a smaller number of examples exhibiting low‐to‐high (“inverse”) dispersion, which to our knowledge has not been reported before. Motivated by a recent report of regular electron dispersion produced by auroral rays moving faster than the E × B drift speed, we investigate a heuristic model of electron acceleration within a region of uniform electric field parallel to B which extends a distance La along magnetic field lines. We show that in addition to a broad range of energies, this model produces inverse dispersion when the detector is less than La beneath the bottom of the acceleration region and regular dispersion for detector distances larger than La. This simple model is meant to inform future efforts to construct a more physical model of suprathermal electron acceleration within moving auroral forms and suggests that inverse dispersion indicates relative proximity to an altitude‐extended acceleration region.

Simple t-designs: a recursive construction for arbitrary t

The aim of this paper is to present a recursive construction of simple t-designs for arbitrary t. The construction is of purely combinatorial nature and it requires finding solutions for the indices of the ingredient designs that satisfy a certain set of equalities. We give a small number of examples to illustrate the construction, whereby we have found a large number of new t-designs, which were previously unknown. This indicates that the method is useful and powerful.

MapReduce program synthesis

By abstracting away the complexity of distributed systems, large-scale data processing platforms—MapReduce, Hadoop, Spark, Dryad, etc.—have provided developers with simple means for harnessing the power of the cloud. In this paper, we ask whether we can automatically synthesize MapReduce-style distributed programs from input–output examples. Our ultimate goal is to enable end users to specify large-scale data analyses through the simple interface of examples. We thus present a new algorithm and tool for synthesizing programs composed of efficient data-parallel operations that can execute on cloud computing infrastructure. We evaluate our tool on a range of real-world big-data analysis tasks and general computations. Our results demonstrate the efficiency of our approach and the small number of examples it requires to synthesize correct, scalable programs.

A new transfer learning framework with application to model-agnostic multi-task learning

Learning from small number of examples is a challenging problem in machine learning. An effective way to improve the performance is through exploiting knowledge from other related tasks. Multi-task...

Cell scale host-pathogen modeling: another branch in the evolution of constraint-based methods.

Constraint-based models have become popular methods for systems biology as they enable the integration of complex, disparate datasets in a biologically cohesive framework that also supports the description of biological processes in terms of basic physicochemical constraints and relationships. The scope, scale, and application of genome scale models have grown from single cell bacteria to multi-cellular interaction modeling; host-pathogen modeling represents one of these examples at the current horizon of constraint-based methods. There are now a small number of examples of host-pathogen constraint-based models in the literature, however there has not yet been a definitive description of the methodology required for the functional integration of genome scale models in order to generate simulation capable host-pathogen models. Herein we outline a systematic procedure to produce functional host-pathogen models, highlighting steps which require debugging and iterative revisions in order to successfully build a functional model. The construction of such models will enable the exploration of host-pathogen interactions by leveraging the growing wealth of omic data in order to better understand mechanism of infection and identify novel therapeutic strategies.

Online visualization of brain connectivity

BackgroundWhile visualization of brain activity has well established practical applications such as real-time functional mapping or neurofeedback, visual representation of brain connectivity is not widely used. In addition, technically challenging single-trial connectivity estimation may have hindered practical usage of connectivity in online applications. New methodIn this work, we developed algorithms that are capable of estimating and visualizing (effective) connectivity between independent cortical sources during online EEG recordings. ResultsThe core routines of our procedure, such as CSPVARICA source extraction and regularized connectivity estimation, are available in our open source Python-based toolbox SCoT. We demonstrate for the first time that online connectivity visualization is feasible. We show this in a feasibility study with twelve participants performing two different tasks, namely motor execution and resting with eyes open or closed. Connectivity patterns were significantly different between two motor tasks in four participants, whereas significant differences between resting task patterns were found in seven participants. Comparison with existing methodsExisting connectivity studies have focused on offline methods. In contrast, there are only a small number of examples in the literature that explored online connectivity estimation. For example, a system based on wearable EEG has been demonstrated to work for one subject, and the Glass Brain project has received considerable attention in popular sciences last year. However, none of these attempts validate their methods on multiple subjects. ConclusionsOur results show that causal connectivity patterns can be observed online during EEG measurements, which is a first step towards real-time connectivity analysis.