Abstract: An apparatus and a method are provided for learning a program with a large number of parameters. In one embodiment, a method not only distorts the input values, but also distorts some of the parameters in the program model. Such an approach not only forces the learned program to acquire parameter values to predict missing or desired data, but also to correct errors in the input data and the program parameters themselves, thereby rendering the learned program more resilient to overfitting and falling into local optima.

Abstract: An apparatus and a method are provided for learning a program with a large number of parameters. In one embodiment, a method not only distorts the input values, but also distorts some of the parameters in the program model. Such an approach not only forces the learned program to acquire parameter values to predict missing or desired data, but also to correct errors in the input data and the program parameters themselves, thereby rendering the learned program more resilient to overfitting and falling into local optima.

Abstract: A method for verifying an identity of a user includes (a) receiving a request for an identification phrase from a user associated with a user account; (b) selecting the phrase from a pool of phrases and associating the phrase with the user account; (c) sending the phrase to the user; (d) receiving a video recording showing a person reading one or more phrases aloud, and (e) verifying that a phrase read by the person in the video recording corresponds to the selected phrase. A third party may desire verification of the user's identity. The method confirms to the third party the user's identity based on whether or not the phrase read aloud in the video recording matches the randomly selected phrase associated with the user account.

Abstract: An apparatus and a method are provided for learning a program with a large number of parameters. In one embodiment, a method not only distorts the input values, but also distorts some of the parameters in the program model. Such an approach not only forces the learned program to acquire parameter values to predict missing or desired data, but also to correct errors in the input data and the program parameters themselves, thereby rendering the learned program more resilient to overfitting and falling into local optima.

Abstract: An optimization in machine learning is achieved using Newton's algorithm together with an efficient technique for solving linear equations, such as the method of conjugate residuals. The techniques of the present invention are applicable to learning language models, predicting classes of objects from images and videos, and classifying financial transactions for prevention of fraud. Other uses include determining a function from a sequence of words to a relevant web page for a search engine, or to inverting arbitrary output values of an analyzed system into an internally running simulation.

Abstract: A method for verifying an identity of a user includes (a) receiving a request for an identification phrase from a user associated with a user account; (b) selecting the phrase from a pool of phrases and associating the phrase with the user account; (c) sending the phrase to the user; (d) receiving a video recording showing a person reading one or more phrases aloud, and (e) verifying that a phrase read by the person in the video recording corresponds to the selected phrase. A third party may desire verification of the user's identity. The method confirms to the third party the user's identity based on whether or not the phrase read aloud in the video recording matches the randomly selected phrase associated with the user account.

Abstract: A method and a system allow accessing several of a user's controlled access accounts by presenting the credentials of only one of the accounts. The method may include (a) storing the credentials for each of the user's accounts; (b) receiving from the user credentials corresponding to any of the user's accounts; (c) presenting the received credentials to access the corresponding account; and (d) upon successful access of the corresponding account, using the stored credentials to access one or more of the user's accounts without requiring the user to present the corresponding credentials. For each of the user's accounts, the credentials are stored encrypted, using a randomly generated key, common to all the encrypted credentials. In addition, the randomly generated key is encrypted using the credentials of each of the accounts. In that manner, plain-text copies of neither the random key nor the credentials of the accounts need to be stored.

Abstract: A system and a method provide in memory two level of indices to a data set, respectively referred to as an instance layer and an entry layer. The instance layer is brought from disk into memory when an application is initialized to allow insertion of new data into the data set. A temporary data structure augments the instance layer and is provided to keep track of the new data. When query processing is required, the entry layer is brought in, so that query processing may be achieved using both the instance layer indices and the entry layer indices. The temporary data is merged into the instance and entry layers and the data set is updated upon termination of the application or at predetermined times, whereupon the resulting data set and index layers are written out to disk.

Abstract: A method allows participants of a social medium or network to unrestricted communication with each other (i.e., without first requiring a pre-established connection or relationship between the communicating participants), without an unacceptable risk of unwanted or unwelcomed communications.

Abstract: A method for announcing arrival of an electronic communication in a wireless mobile device includes: (a) receiving from a wireless interface of the mobile device data corresponding to the arrival of the electronic communication; (b) processing the received data to extract therefrom identification of information, the identification information including an identity of a sender or caller; (c) encoding the extracted identification in an encoding scheme suitable for being played physically in an actuator of the mobile device; and (d) providing the encoded identification to a driver for the actuator of the mobile device. The identification information may further include a subject of the electronic communication. The encoding scheme may be, for example, a form of Morse code. The encoded identification may therefore be expressed by an alphabet represented by segments of audible tone of various durations to be played by the audio system of the mobile device.

Abstract: A method and a system use an intermediate server to process the communication between two parties, so as to eliminate echoes between them. The server performs echo cancellation in a network-based voice communication system handling a large number of conversations. In one implementation, the server allocates two echo cancellation modules to each conversation, with each echo cancellation module including (a) a communication interface for communicating with a client program associated with the echo cancellation module; (b) a first buffer for storing audio data received from the client program for transmission to another echo cancellation module; (c) a second buffer for storing audio data received from the other echo cancellation module for transmitting to the associated client program; and (d) a set of filters using the audio data in both the first buffer and the second buffer to cancel echoes in the audio data in the second buffer.

Abstract: Methods and apparatus, including computer program products, for identifying Web page content with a granularity finer than individual Web pages, e.g., finer than individual HTML documents. The invention provides a computer-implemented method for identifying Web page content. The method includes receiving a string of markup language source code that includes tags. The method includes identifying sub-sequences in which tags occur in the string. Each sub-sequence is associated with the portion of the string that starts with the first tag of the sub-sequence and ends with the last tag of the sub-sequence. The sub-sequences identified are ones that satisfy criteria for being classified as associated with a portion of the string that define Web page content constituting an entire listing. The criteria includes a requirement that an identified sub-sequence be repeated in tandem, either exactly or approximately, in the string. The method includes returning the identified sub-sequences.