Learning Kernels for Unsupervised Domain Adaptation with Applications to Visual Object Recognition

Domain adaptation aims to correct the mismatch in statistical properties between the source domain on which a classifier is trained and the target domain to which the classifier is to be applied. In this paper, we address the challenging scenario of unsupervised domain adaptation, where the target domain does not provide any annotated data to assist in adapting the classifier. Our strategy is to learn robust features which are resilient to the mismatch across domains and then use them to construct classifiers that will perform well on the target domain. To this end, we propose novel kernel learning approaches to infer such features for adaptation. Concretely, we explore two closely related directions. In the first direction, we propose unsupervised learning of a geodesic flow kernel (GFK). The GFK summarizes the inner products in an infinite sequence of feature subspaces that smoothly interpolates between the source and target domains. In the second direction, we propose supervised learning of a kernel that discriminatively combines multiple base GFKs. Those base kernels model the source and the target domains at fine-grained granularities. In particular, each base kernel pivots on a different set of landmarks—the most useful data instances that reveal the similarity between the source and the target domains, thus bridging them to achieve adaptation. Our approaches are computationally convenient, automatically infer important hyper-parameters, and are capable of learning features and classifiers discriminatively without demanding labeled data from the target domain. In extensive empirical studies on standard benchmark recognition datasets, our appraches yield state-of-the-art results compared to a variety of competing methods.     

Design Principles for Cationic,Astrocyte-Targeted Probes

The brain's astrocytes play key roles in normal and pathological brain processes. Targeting small molecules to astrocytes in the presence of the many other cell types in the brain will provide useful tools for their visualization and manipulation. Herein, we explore the functional consequences of synthetic modifications to a recently described astrocyte marker composed of a bright rhodamine-based fluorophore and an astrocyte-targeting moiety. We altered the nature of the targeting moiety to probe the dependence of astrocyte targeting on hydrophobicity, charge, and pK_a when exposed to astrocytes and neurons isolated from the mouse cortex. We found that an overall molecular charge of +2 and a targeting moiety with a heterocyclic aromatic amine are important requirements for specific and robust astrocyte labeling. These results provide a basis for engineering astrocyte-targeted molecular tools with unique properties, including metabolite sensing or optogenetic control.     

Large-Scale Live Active Learning: Training Object Detectors with Crawled Data and Crowds

Active learning and crowdsourcing are promising ways to efficiently build up training sets for object recognition, but thus far techniques are tested in artificially controlled settings. Typically the vision researcher has already determined the dataset’s scope, the labels “actively” obtained are in fact already known, and/or the crowd-sourced collection process is iteratively fine-tuned. We present an approach for live learning of object detectors, in which the system autonomously refines its models by actively requesting crowd-sourced annotations on images crawled from the Web. To address the technical issues such a large-scale system entails, we introduce a novel part-based detector amenable to linear classifiers, and show how to identify its most uncertain instances in sub-linear time with a hashing-based solution. We demonstrate the approach with experiments of unprecedented scale and autonomy, and show it successfully improves the state-of-the-art for the most challenging objects in the PASCAL VOC benchmark. In addition, we show our detector competes well with popular nonlinear classifiers that are much more expensive to train.     

Oxidative Stability of Chia (Salvia hispanica L.) and Sesame (Sesamum indicum L.) Oil Blends

Chia and sesame oils are important sources of essential fatty acids; however, their ω-3:ω-6 proportions do not comply with nutritional recommendation. A feasible approach to improve the ratio is to blend different oils, but only after understanding physical and chemical changes of the new matrix. Objective of the investigation was to determine the physico-chemical characteristics and the oxidative stability index (OSI), using the Rancimat method, of chia-sesame oil blends. The four ω-3:ω-6 blends tested (1:4, 1:6, 1:8, and 1:10) were exposed to temperatures of 110, 120, and 130 °C. The OSI values of the mixtures varied between 6.24–8.08, 3.07–4.00, and 1.62–2.01 hours for each temperature, respectively. In addition, their mean activation energy, enthalpy, entropy, and Q₁₀ were 88.4 kJ/mol, 85.2 kJ/mol, −41.1 J/mol K, and 2.0. Finally, a shelf life prediction performed at 25 °C indicated stability times between 80 and 123 days. Therefore, combining chia and sesame oils produced blends with a good balance of essential fatty acids.     

In Silico/In Vitro Hit-to-Lead Methodology Yields SMYD3 Inhibitor That Eliminates Unrestrained Proliferation of Breast Carcinoma Cells

SMYD3 is a lysine methyltransferase that regulates the expression of over 80 genes and is required for the uncontrolled proliferation of most breast, colorectal, and hepatocellular carcinomas. The elimination of SMYD3 restores normal expression patterns of these genes and halts aberrant cell proliferation, making it a promising target for small molecule inhibition. In this study, we sought to establish a proof of concept for our in silico/in vitro hit-to-lead enzyme inhibitor development platform and to identify a lead small molecule candidate for SMYD3 inhibition. We used Schrodinger^® software to screen libraries of small molecules in silico and the five compounds with the greatest predicted binding affinity within the SMYD3 binding pocket were purchased and assessed in vitro in direct binding assays and in breast cancer cell lines. We have confirmed the ability of one of these inhibitors, Inhibitor-4, to restore normal rates of cell proliferation, arrest the cell cycle, and induce apoptosis in breast cancer cells without affecting wildtype cell behavior. Our results provide a proof of concept for this fast and affordable small molecule hit-to-lead methodology as well as a promising candidate small molecule SMYD3 inhibitor for the treatment of human cancer.     

Contact conductivity detection in poly(methyl methacrylate)-based microfluidic devices for analysis of mono- and polyanionic molecules

An on-column contact conductivity detector was developed for the analysis of various mono- and polyanionic compounds separated by electrophoresis chips fabricated in poly(methyl methacrylate) (PMMA) using hot embossing techniques from Ni electroforms. The detector consisted of a pair of Pt wires (127 microm diameter) with an end-to-end spacing of approximately 20 microm and situated within the fluidic channel. The waveform applied to the electrode pair was a bipolar pulse with a frequency of 5.0 kHz and was used to reduce the charging current from measurement so that the current recorded at the end of one pulse is more representative of the solution conductivity. Using the detector, separations of amino acids, peptides, proteins, and oligonucleotides were demonstrated. For the amino acids and peptides, free-solution zone electrophoresis was performed. A calibration plot for the amino acid alanine was found to be linear from approximately 10 to 100 nM in a carrier electrolyte consisting of 10 mM triethylamonium acetate. The concentration detection limit was found to be 8.0 nM, with the corresponding mass detection limit equal to 3.4 amol (injection volume = 425 pL). The protein separations with conductivity detection were performed using MEKC, in which the carrier electrolyte contained the anionic surfactant sodium dodecyl sulfate (SDS) above its cmc. Near baseline resolution was achieved in the PMMA microchip for a solution containing 8 different proteins. In the case of the DNA fragments, capillary electrochromatography was used with a C18-modified PMMA chip and a carrier electrolyte containing an ion-pairing agent.     

The use of background and ability profiles to predict college student outcomes.

To determine whether profiles of predictor variables provide incremental prediction of college student outcomes, the authors 1st applied an empirical clustering method to profiles based on the scores of 2,771 entering college students on a battery of biographical data and situational judgment measures, along with SAT and American College Test scores and high school grade point average, which resulted in 5 student groups. Performance of the students in these clusters was meaningfully different on a set of external variables, including college grade point average, self-rated performance, class absenteeism, organizational citizenship behavior, intent to quit their university, and satisfaction with college. The 14 variables in the profile were all significantly correlated with 1 or more of the outcome measures; however, nonlinear prediction of these outcomes on the basis of cluster membership did not add incrementally to a linear-regression-based combination of these 14 variables as predictors. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Protein,ash, lutein and tocols distribution in einkorn (Triticum monococcum L. subsp. monococcum) seed fractions

The distribution of protein, ash, lutein, tocopherols and tocotrienols in the germ, bran and endosperm portions was studied in seeds of two einkorn accessions and one bread wheat. The two einkorns showed a higher content of most compounds, but the distribution within the kernel was similar in both species. The germ fraction showed the highest concentration of protein, lutein, α-tocopherol, β-tocopherol and total tocols. Ash, α-tocotrienol and β-tocotrienol levels were highest in the bran fraction, although significant quantities were detected also in the germ and, for tocotrienols, in the flour.     

A tool for power and phase noise optimization in frequency synthesizers

In this paper, we present a CAD technique to design low-power and low phase noise integrated frequency synthesizers. This technique introduces a key parameter, Phase Noise per Unit Power, which correlates phase noise and power among all the sub-circuits in the frequency synthesizer. By correlating the performance of all the independent circuits together, sophisticated synthesizer design and optimization can be significantly simplified. We demonstrate a 1.8 GHz frequency synthesizer design in a 0.18 μm CMOS process achieving −132 dBc/Hz phase noise at 100 kHz offset with less than 4.3 mW power consumption.