Abstract: In a neural network, hidden layers are modified by supplying input data, an output label, and internal teaching labels to the neural network; causing the neural network to process the input data through the hidden layers and outputting a result of the processing for comparison with the output label; supplying the internal teaching labels to the hidden layers and calculating scores for the hidden layers based on the internal teaching labels; and modifying the hidden layers or hidden nodes based on the calculated scores and the comparison of the processing result with the output label. The modifications to the hidden layers or hidden nodes may involve pruning hidden nodes by dropping lower scoring nodes; reducing a number of bits in computations and outputs; reducing a number of bits in selected nodes; bypassing lower scoring nodes; modifying activation functions of the hidden nodes based on the calculated scores; and/or adding hidden layers or hidden nodes.

Abstract: A method for antenna subset selection by joint processing in RF and baseband in a multi-antenna systems. Lt input data streams are generated in a transmitter for either diversity transmission or multiplexing transmission. These streams are modulated to RF signals. These signals are switched to the t branches associated with the t transmit antennas, and a phase-shift transformation is applied to the RF signals by a t×t matrix multiplication operator ?1, whose output are t?Lt RF signals. These signals are transmitted over a channel by t antennas. The transmitted signals are received by r antennas in a receiver. A phase-shift transformation is applied to the r RF signals by a r×r matrix multiplication operator ?2. Lr branches of these phase shifted streams are demodulated and further processed in baseband to recover the input data streams.

Abstract: A source signal transmitted through multiple channels having space, time and frequency diversities to generate multiple received signals is recovered by an iterative associative memory model with dynamic maximum likelihood estimation. A current symbol vector representing the multiple received signals is projected to a net-vector using a linear matrix operation with a weight matrix W. The weight matrix is obtained by a singular value decomposition of an input symbol sequence. The net-vector is mapped to a nearest symbol vector using a non-linear operation with an activation function. The projecting and mapping steps are repeated until the nearest symbol vector converges to a valid symbol vector representing the source signal.

Abstract: A method for dynamically allocating network resources while transferring multimedia at variable bit-rates in a network extracts first content features from the multimedia to determine renegotiation points and observation periods. Second content features and traffic features are extracted from the multimedia bit stream during the observation periods. The second content features and the traffic features are combined in a neural network to predict the network resources to be allocated at the renegotiation points.

Abstract: A method transmits data streams in a multiple-input/multiple-output (MIMO) wireless communication systems, where the number of receiving antennas q is less than a number of the transmitting antennas p. The data streams are precoded with a set of finite impulse response filters according to a transfer function of the MIMO channels. The precoded data streams are transmitted over multiple-input/multiple-output channels to a receiver, where the transmitted precoded data stream are detected and decoded to perfectly recover the plurality of data streams without the use of an equalizer.

Abstract: A receiver in a multiple-input-multiple-output, frequency-selective fading wireless communication systems sequentially recovers multiple data stream. A next input stream, having a highest signal-to-noise ratio is selected. The selected input stream is equalized, detected and decoded. The decoded data stream is then substracted from the data streams, and the selecting, equalizing, detecting and decoding, and subtracting is repeated until all of the data streams have been decoded.

Abstract: A method for generating a representation of multimedia content by first segmenting the multimedia content spatially and temporally to extract objects. Feature extraction is applied to the objects to produce semantic and syntactic attributes, relations, and a containment set of content entities. The content entities are coded to produce directed acyclic graphs of the content entities, where each directed acyclic graph represents a particular interpretation of the multimedia content.

Abstract: A source signal transmitted through multiple channels having space, time and frequency diversities to generate multiple received signals is recovered by an iterative associative memory model with dynamic maximum likelihood estimation. A current symbol vector representing the multiple received signals is projected to a net-vector using a linear matrix operation with a weight matrix W. The weight matrix is obtained by a singular value decomposition of an input symbol sequence. The net-vector is mapped to a nearest symbol vector using a non-linear operation with an activation function The projecting and converting steps are repeated until the nearest symbol vector converges to a valid symbol vector representing the source signal.

Abstract: A method manages traffic over a channel of a network connecting a sender end system and a receiver end system. The traffic includes multimedia packets. The channel is modeled as a queue having an associated queue occupancy. The times when packets are sent and the times when feedback messages are received are maintained in the sender end system. A time series of samples for a service time experienced by each packet sent is updated based on the total number of packets sent and the total number of feedback messages received. A queue occupancy for a next packet to be sent is then predicted based on the time series, and the next packet is sent according to the predicted queue occupancy.

Abstract: A method for dynamically allocating network resources while transferring multimedia at variable bit-rates in a network extracts first content features from the multimedia to determine renegotiation points and observation periods. Second content features and traffic features are extracted from the multimedia bit stream during the observation periods. The second content features and the traffic features are combined in a neural network to predict the network resources to be allocated at the renegotiation points.

Abstract: A method determines a surface of an object in a sequence of images. The method begins by estimating a boundary of the object in each image of the sequence using motion information of adjacent images of the sequence. Then, portions of each image of the sequence are ordered to produce an ordered sequence of images. The ordered portions are exterior to the estimated object boundary. Edges in each ordered image are filtered using the motion information, and each ordered image of the sequence is searched to locate the filtered edges to form a new boundary outside the estimated boundary. The filtering and searching are repeated, while projecting the new object boundaries over the sequence of images, until the new object boundaries converges to a surface of the object.

Abstract: A technique is provided for calculating the time offsets between different video cameras and re-synchronizing the captured frames in a post-processing manner, thus eliminating the necessity of an explicit common clock for synchronization. This approach allows effective synchronization of frames from different cameras so that a multi-camera system can be used to more accurately analyze a subject under observation.

Abstract: A method represents a shape of an object in an image. Portions of the image interior to a boundary of the object are ordered. A medial axis for the ordered portions of the image is derived. The medial axis is segmented to produce an ordered tree. The ordered tree is transformed to a partial ordered tree to represent the shape of the object.

Abstract: A method detects a boundary in a sequence of two-dimensional images where each image has multiple intensity value points. Filtering and motion analysis is applied on each image to produce motion enhanced images. Initial search parameters are determined from a dynamic snake model applied to the motion enhanced images. Each motion enhanced image is searched for a potential boundary using the search parameters. The potential boundary is projected into the motion enhanced image of a previous, current, and next image, and the search parameters of the previous, current, and next images are updated. The searching, projecting, and updating repeat until a predetermined level of convergence is reached.

Abstract: A method determines true motion vectors associated with a sequence of images. The images include fields made up of blocks of pixels. The method selects candidate feature blocks from the blocks of pixels. The candidate feature blocks have intensity variances above a threshold indicative of texture features. Candidate feature blocks in similarly numbered adjacent field intervals are compared to determine sets of displaced frame differences parameters for each candidate feature block. The true motion vectors for each candidate feature block are determined from a minimum weighted score derived from the difference parameters.

Abstract: A system for automatically detecting and recognizing the identity of a deformable object such as a human face, within an arbitrary image scene. The system comprises an object detector implemented as a probabilistic DBNN, for determining whether the object is within the arbitrary image scene and a feature localizer also implemented as a probabilistic DBNN, for determining the position of an identifying feature on the object such as the eyes. A feature extractor is coupled to the feature localizer and receives coordinates sent from the feature localizer which are indicative of the position of the identifying feature and also extracts from the coordinates information relating to other features of the object such as the eyebrows and nose, which are used to create a low resolution image of the object. A probabilistic DBNN based object recognizer for determining the identity of the object receives the low resolution image of the object inputted from the feature extractor to identify the object.