Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: Systems and methods are provided to pre-train projection networks for use as transferable natural language representation generators. In particular, example pre-training schemes described herein enable learning of transferable deep neural projection representations over randomized locality sensitive hashing (LSH) projections, thereby surmounting the need to store any embedding matrices because the projections can be dynamically computed at inference time.

Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: A method, implemented on at least one computing device each of which has at least one processor, storage, and a communication platform connected to a network for providing synthetic answers to a personal question is disclosed. A personal question is received from a person. One or more entities are extracted from the personal question. One or more relations are extracted from the personal question. A model is selected based on the personal question. One or more synthetic answers to the personal question are obtained based on the one or more entities, the one or more relations, and the selected model.

Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSegoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSegoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: Systems and methods for generating human readable natural language summary for campaign audience are provided. The system includes a memory storing a database including audience segments and tags related to the audience segments. A computer server is in communication with the memory and the database, the computer server programmed to: obtain campaign delivery feed data related to a plurality of campaigns from at least one advertiser in a preset time period; obtain audience feed data including tag information from a data provider; cluster the tag information to find term frequencies for each term in the tag information; identify human understandable terms from the clustered tag information by removing noisy terms; and generate a human understandable report using the human understandable terms in a timely fashion.

Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: Systems and methods are provided to pre-train projection networks for use as transferable natural language representation generators. In particular, example pre-training schemes described herein enable learning of transferable deep neural projection representations over randomized locality sensitive hashing (LSH) projections, thereby surmounting the need to store any embedding matrices because the projections can be dynamically computed at inference time.

Abstract: A method, implemented on at least one computing device, each of which has at least one processor, storage, and a communication platform connected to a network for classifying a question is disclosed. A question is received from a person. A question pattern is determined. A model selected based on the question is retrieved. Further, a decision is made as to whether the question is a personal question based on the question pattern and the selected model.

Abstract: The present disclosure provides projection neural networks and example applications thereof. In particular, the present disclosure provides a number of different architectures for projection neural networks, including two example architectures which can be referred to as: Self-Governing Neural Networks (SGNNs) and Projection Sequence Networks (ProSeqoNets). Each projection neural network can include one or more projection layers that project an input into a different space. For example, each projection layer can use a set of projection functions to project the input into a bit-space, thereby greatly reducing the dimensionality of the input and enabling computation with lower resource usage. As such, the projection neural networks provided herein are highly useful for on-device inference in resource-constrained devices. For example, the provided SGNN and ProSeqoNet architectures are particularly beneficial for on-device inference such as, for example, solving natural language understanding tasks on-device.

Abstract: The present teaching relates to updating an intent space and estimating intent based on an intent space. In one example, an initial intent space is obtained. Each intent in the initial intent space is characterized in one or more dimensions. At least one model is received. Each of the at least one model provides features in each of the dimensions and relationship thereof. A new intent associated with an intent in the initial intent space is determined based on the at least one model. Based on the new intent, the initial intent space is updated to derive an updated intent space.

Abstract: Systems and methods for generating human readable natural language summary for campaign audience are provided. The system includes a memory storing a database including audience segments and tags related to the audience segments. A computer server is in communication with the memory and the database, the computer server programmed to: obtain campaign delivery feed data related to a plurality of campaigns from at least one advertiser in a preset time period; obtain audience feed data including tag information from a data provider; cluster the tag information to find term frequencies for each term in the tag information; identify human understandable terms from the clustered tag information by removing noisy terms; and generate a human understandable report using the human understandable terms in a timely fashion.

Abstract: A method, implemented on at least one computing device, each of which has at least one processor, storage, and a communication platform connected to a network for classifying a question is disclosed. A question is received from a person. A question pattern is determined. A model selected based on the question is retrieved. Further, a decision is made as to whether the question is a personal question based on the question pattern and the selected model.

Abstract: A method, implemented on at least one computing device, each of which has at least one processor, storage, and a communication platform connected to a network for classifying a question is disclosed. A question is received from a person. A question pattern is determined. A model selected based on the question is retrieved. Further, a decision is made as to whether the question is a personal question based on the question pattern and the selected model.

Abstract: The present teaching relates to generating a card based on intent. In one example, a request is received for generating a card to be provided to a user. Intent of the user with respect to the card is obtained. One or more modules are selected to be put into the card based on the intent. The card is generated based on the selected one or more modules.

Abstract: A method, implemented on at least one computing device each of which has at least one processor, storage, and a communication platform connected to a network for providing synthetic answers to a personal question is disclosed. A personal question is received from a person. One or more entities are extracted from the personal question. One or more relations are extracted from the personal question. A model is selected based on the personal question. One or more synthetic answers to the personal question are obtained based on the one or more entities, the one or more relations, and the selected model.

Abstract: A method, implemented on at least one computing device, each of which has at least one processor, storage, and a communication platform connected to a network for classifying a question is disclosed. A question is received from a person. A question pattern is determined. A model selected based on the question is retrieved. Further, a decision is made as to whether the question is a personal question based on the question pattern and the selected model.

Abstract: The present teaching relates to updating an intent space and estimating intent based on an intent space. In one example, an initial intent space is obtained. Each intent in the initial intent space is characterized in one or more dimensions. At least one model is received. Each of the at least one model provides features in each of the dimensions and relationship thereof. A new intent associated with an intent in the initial intent space is determined based on the at least one model. Based on the new intent, the initial intent space is updated to derive an updated intent space.

Abstract: The present teaching relates to generating a card based on intent. In one example, a request is received for generating a card to be provided to a user. Intent of the user with respect to the card is obtained. One or more modules are selected to be put into the card based on the intent. The card is generated based on the selected one or more modules.