Fast support vector machines for convolution tree kernels

Feature engineering is one of the most complex aspects of system design in machine learning. Fortunately, kernel methods provide the designer with formidable tools to tackle such complexity. Among others, tree kernels (TKs) have been successfully applied for representing structured data in diverse domains, ranging from bioinformatics and data mining to natural language processing. One drawback of such methods is that learning with them typically requires a large number of kernel computations (quadratic in the number of training examples) between training examples. However, in practice substructures often repeat in the data which makes it possible to avoid a large number of redundant kernel evaluations. In this paper, we propose the use of Directed Acyclic Graphs (DAGs) to compactly represent trees in the training algorithm of Support Vector Machines. In particular, we use DAGs for each iteration of the cutting plane algorithm (CPA) to encode the model composed by a set of trees. This enables DAG kernels to efficiently evaluate TKs between the current model and a given training tree. Consequently, the amount of total computation is reduced by avoiding redundant evaluations over shared substructures. We provide theory and algorithms to formally characterize the above idea, which we tested on several datasets. The empirical results confirm the benefits of the approach in terms of significant speedups over previous state-of-the-art methods. In addition, we propose an alternative sampling strategy within the CPA to address the class-imbalance problem, which coupled with fast learning methods provides a viable TK learning framework for a large class of real-world applications.     

Resolution proof transformation for compression and interpolation

Verification methods based on SAT, SMT, and theorem proving often rely on proofs of unsatisfiability as a powerful tool to extract information in order to reduce the overall effort. For example a proof may be traversed to identify a minimal reason that led to unsatisfiability, for computing abstractions, or for deriving Craig interpolants. In this paper we focus on two important aspects that concern efficient handling of proofs of unsatisfiability: compression and manipulation. First of all, since the proof size can be very large in general (exponential in the size of the input problem), it is indeed beneficial to adopt techniques to compress it for further processing. Secondly, proofs can be manipulated as a flexible preprocessing step in preparation for interpolant computation. Both these techniques are implemented in a framework that makes use of local rewriting rules to transform the proofs. We show that a careful use of the rules, combined with existing algorithms, can result in an effective simplification of the original proofs. We have evaluated several heuristics on a wide range of unsatisfiable problems deriving from SAT and SMT test cases.     

A low-power wireless video sensor node for distributed object detection

In this paper we propose MicrelEye, a wireless video node for cooperative distributed video processing applications that involve image classification. The node is equipped with a low-cost VGA CMOS image sensor, a reconfigurable processing engine (FPGA, Microcontroller, SRAM) and a Bluetooth 100-m transceiver. It has a size of few cubic centimeters and its typical power consumption is approximately ten times less than that of typical commercial DSP-based solutions. As regards classification, a highly optimized hardware-oriented support vector machine-like (SVM-like) algorithm called ERSVM is proposed and implemented. We describe our hardware and software architecture, its performance and power characteristics. The case study considered in this paper is people detection. The obtained results suggest that the present technology allows for the design of simple intelligent video nodes capable of performing classification tasks locally.

Numerical simulations of phenomena observed at hot microelectrodes: Inlaid microdisk

In this paper we describe numerical simulations of a number of physico-chemical phenomena observed at disk microelectrodes polarized with a high frequency ac waveform. These phenomena include resistive heating of an electrolyte solution surrounding a microelectrode, electrothermal flow of the solution and electrochemical mass transport of dissolved redox species. The main purpose of doing these simulations was to provide the required theoretical background to the previously obtained experimental data. It is shown that the simulated linear flow rate of the solution is proportional to the voltage across the solution resistance in the power four and inversely proportional to the radius of the microelectrode. Thus this supports the idea that the convection at hot microelectrodes is driven primarily by the electrothermal flow. Simulations of the Faradaic current agree semi-quantitatively with the experimental observations. We have identified possible reasons for the observed discrepancy between the two sets of data.     

Liquid–Liquid Diffusion‐Assisted Crystallization: A Fast and Versatile Approach Toward High Quality Mixed Quantum Dot‐Salt Crystals

Here, a new, fast, and versatile method for the incorporation of colloidal quantum dots (QDs) into ionic matrices enabled by liquid–liquid diffusion is demonstrated. QDs bear a huge potential for numerous applications thanks to their unique chemical and physical properties. However, stability and processability are essential for their successful use in these applications. Incorporating QDs into a tight and chemically robust ionic matrix is one possible approach to increase both their stability and processability. With the proposed liquid–liquid diffusion‐assisted crystallization (LLDC), substantially accelerated ionic crystallization of the QDs is shown, reducing the crystallization time needed by one order of magnitude. This fast process allows to incorporate even the less stable colloids including initially oil‐based ligand‐exchanged QDs into salt matrices. Furthermore, in a modified two‐step approach, the seed‐mediated LLDC provides the ability to incorporate oil‐based QDs directly into ionic matrices without a prior phase transfer. Finally, making use of their processability, a proof‐of‐concept white light emitting diode with LLDC‐based mixed QD‐salt films as an excellent color‐conversion layer is demonstrated. These findings suggest that the LLDC offers a robust, adaptable, and rapid technique for obtaining high quality QD‐salts.     

Dielectrophoretic and electrothermal effects at alternating current heated disk microelectrodes

When a disk microelectrode is polarized with an alternating potential of very high frequency (0.1-2 GHz) and a high amplitude (up to 2.8 V rms), the electrode is heated up, and at the same time, a very intense electric field is created around the electrode (>10(6) V/m for electrodes 1 microm in radius). This strong electric field gives rise to positive or negative dielectrophoretic effects. Positive dielectrophoretic effects can be used to assemble nanowires from nanoparticles at the electrode edge. On the other hand, a negative dielectrophoretic effect is probably responsible for "jet boiling" observed at overheated microelectrodes. In addition, a combination of a high temperature gradient and a high potential gradient generates an intense electrothermal flow of solution which very strongly enhances the mass transport and is responsible for intense convection in such systems. The electrothermal flow and dielectrophoretic forces can be generated directly on a microelectrode employed in electrochemical detection because the high frequency ac polarization of the electrode does not interfere with the acquisition of analytical signals.     

An abstraction refinement approach combining precise and approximated techniques

Predicate abstraction is a powerful technique to reduce the state space of a program to a finite and affordable number of states. It produces a conservative over-approximation where concrete states are grouped together according to a given set of predicates. A precise abstraction contains the minimal set of transitions with regard to the predicates, but as a result is computationally expensive. Most model checkers therefore approximate the abstraction to alleviate the computation of the abstract system by trading off precision with cost. However, approximation results in a higher number of refinement iterations, since it can produce more false counterexamples than its precise counterpart. The refinement loop can become prohibitively expensive for large programs. This paper proposes a new approach that employs both precise (slow) and approximated (fast) abstraction techniques within one abstraction-refinement loop. It allows computing the abstraction quickly, but keeps it precise enough to avoid too many refinement iterations. We implemented the new algorithm in a state-of-the-art software model checker. Our tests with various real-life benchmarks show that the new approach almost systematically outperforms both precise and imprecise techniques.     

Time- and Spectrally-Resolved Photoluminescence Study of Alloyed CdxZn1−xSeyS1−y/ZnS Quantum Dots and Their Nanocomposites with SPIONs in Living Cells

Magnetic-luminescent composites based on semiconductor quantum dots (QDs) and superparamagnetic iron oxide nanoparticles (SPIONs) can serve as a platform combining visualization and therapy. Here, we report the construction of QD-SPION nanocomposites based on synthesized SPIONs and alloyed QDs (CdxZn1−xSeyS1−y)/ZnS solubilized with L-cysteine molecules. The study of the spectral-luminescence characteristics, the kinetics of luminescence decay show the composite’s stability in a solution. After incubation with HeLa cells, QDs, SPIONs, and their composites form clusters on the cell surface and associate with endosomes inside the cells. Component-wise analysis of the photoluminescence decay of cell-associated QDs/SPIONs provides information about their localization and aggregate status.