IBM

About: IBM is a company organization based out in Armonk, New York, United States. It is known for research contribution in the topics: Layer (electronics) & Cache. The organization has 134567 authors who have published 253905 publications receiving 7458795 citations. The organization is also known as: International Business Machines Corporation & Big Blue.

Cost-sensitive learning by cost-proportionate example weighting

Abstract: We propose and evaluate a family of methods for converting classifier learning algorithms and classification theory into cost-sensitive algorithms and theory. The proposed conversion is based on cost-proportionate weighting of the training examples, which can be realized either by feeding the weights to the classification algorithm (as often done in boosting), or by careful subsampling. We give some theoretical performance guarantees on the proposed methods, as well as empirical evidence that they are practical alternatives to existing approaches. In particular, we propose costing, a method based on cost-proportionate rejection sampling and ensemble aggregation, which achieves excellent predictive performance on two publicly available datasets, while drastically reducing the computation required by other methods.

Rotation, scale, and translation resilient watermarking for images

Abstract: Many electronic watermarks for still images and video content are sensitive to geometric distortions. For example, simple rotation, scaling, and/or translation (RST) of an image can prevent blind detection of a public watermark. In this paper, we propose a watermarking algorithm that is robust to RST distortions. The watermark is embedded into a one-dimensional (1-D) signal obtained by taking the Fourier transform of the image, resampling the Fourier magnitudes into log-polar coordinates, and then summing a function of those magnitudes along the log-radius axis. Rotation of the image results in a cyclical shift of the extracted signal. Scaling of the image results in amplification of the extracted signal, and translation of the image has no effect on the extracted signal. We can therefore compensate for rotation with a simple search, and compensate for scaling by using the correlation coefficient as the detection measure. False positive results on a database of 10 000 images are reported. Robustness results on a database of 2000 images are described. It is shown that the watermark is robust to rotation, scale, and translation. In addition, we describe tests examining the watermarks resistance to cropping and JPEG compression.

Low dielectric constant materials.

Abstract: Modern computer microprocessor chips are marvels of engineering complexity. For the current 45 nm technology node, there may be nearly a billion transistors on a chip barely 1 cm2 and more than 10 000 m of wiring connecting and powering these devices distributed over 9-10 wiring levels. This represents quite an advance from the first INTEL 4004B microprocessor chip introduced in 1971 with 10 μm minimum dimensions and 2 300 transistors on the chip! It has been disclosed that advanced microprocessor chips at the 32 nm node will have more than 2 billion transistors.1 For instance, Figure 1 shows a sectional 3D image of a 90 nm IBM microprocessor, containing several hundred million integrated devices and 10 levels of interconnect wiring, designated as the back-end-of-the-line (BEOL). Since the invention of microprocessors, the number of active devices on a chip has been exponentially increasing, approximately doubling every two years. This trend was first described in 1965 by Gordon Moore,2 although the original discussion suggested doubling the number of devices every year, and the phenomenon became popularly known as Moore’s Law. This progress has proven remarkably resilient and has persisted for more than 50 years. The enabler that has permitted these advances is known as scaling, that is, the reduction of minimum device dimensions by lithographic advances (photoresists, tooling, and process integration optimization) by ∼30% for each device generation.3 It allowed more active devices to be incorporated in a given area and improved the operating characteristics of the individual transistors. It should be emphasized that the earlier improvements in chip performance were achieved with very few changes in the materials used in the construction of the chips themselves. The increase of performance with scaling * Corresponding author. E-mail: gdubois@us.ibm.com. † IBM Almaden Research Center. ‡ Stanford University. Willi Volksen received his B.S. in Chemistry (magna cum laude) from New Mexico Institute of Mining and Technology in 1972 and his Ph.D. in Chemistry/Polymer Science from the University of Massachusetts, Lowell, in 1975. He then joined the research group of Prof. Harry Gray/Dr. Alan Rembaum at the California Institute of Technology as a postdoctoral fellow and upon completion of the one-year appointment joined Dr. Rembaum at the Jet Propulsion Laboratory as a Senior Chemist in 1976. In 1977 Dr. Volksen joined the IBM Research Division at the IBM Almaden Research Center in San Jose, CA, where he is an active research staff member in the Advanced Materials Group of the Science and Technology function.

Speaker adaptation of neural network acoustic models using i-vectors

Abstract: We propose to adapt deep neural network (DNN) acoustic models to a target speaker by supplying speaker identity vectors (i-vectors) as input features to the network in parallel with the regular acoustic features for ASR. For both training and test, the i-vector for a given speaker is concatenated to every frame belonging to that speaker and changes across different speakers. Experimental results on a Switchboard 300 hours corpus show that DNNs trained on speaker independent features and i-vectors achieve a 10% relative improvement in word error rate (WER) over networks trained on speaker independent features only. These networks are comparable in performance to DNNs trained on speaker-adapted features (with VTLN and FMLLR) with the advantage that only one decoding pass is needed. Furthermore, networks trained on speaker-adapted features and i-vectors achieve a 5-6% relative improvement in WER after hessian-free sequence training over networks trained on speaker-adapted features only.

Quantum mirages formed by coherent projection of electronic structure

Abstract: Image projection relies on classical wave mechanics and the use of natural or engineered structures such as lenses or resonant cavities. Well-known examples include the bending of light to create mirages in the atmosphere, and the focusing of sound by whispering galleries. However, the observation of analogous phenomena in condensed matter systems is a more recent development, facilitated by advances in nanofabrication. Here we report the projection of the electronic structure surrounding a magnetic Co atom to a remote location on the surface of a Cu crystal; electron partial waves scattered from the real Co atom are coherently refocused to form a spectral image or 'quantum mirage'. The focusing device is an elliptical quantum corral, assembled on the Cu surface. The corral acts as a quantum mechanical resonator, while the two-dimensional Cu surface-state electrons form the projection medium. When placed on the surface, Co atoms display a distinctive spectroscopic signature, known as the many-particle Kondo resonance, which arises from their magnetic moment. By positioning a Co atom at one focus of the ellipse, we detect a strong Kondo signature not only at the atom, but also at the empty focus. This behaviour contrasts with the usual spatially-decreasing response of an electron gas to a localized perturbation.