Abstract: Provided is an acoustic microfluidic system for cell fusion, a preparation method therefor and use thereof, which relates to the technical field of cell fusion. The acoustic microfluidic system of the present invention comprises a signal generator, a power amplifier, a PDMS cavity, a micro-injection pump, a pipeline, an EP tube, a cell recovery container, and a bulk wave transducer/surface acoustic wave transducer. The side wall/bottom of the PDMS cavity is provided with identical microporous structures disposed in a staggered manner The system of the present invention has the advantages of extremely low heat production quantity, simple operation, high repeatability and strong stability, and is suitable for the fusion of homologous cells and non-homologous cells. The system is not only suitable for the fusion of two cells, but also for the fusion of a plurality of cells, and can be widely applied to various types of cells.

Abstract: An optical system (100), sequentially comprising from an object side to an image side: a first lens (L1) having positive refractive power, an object-side surface (S1) of the first lens (L1) being a convex surface at the circumference; a second lens (L2), a third lens (13), a fourth lens (L4), a fifth lens (L5), a sixth lens (L6), and a seventh lens (L7) having refractive power; and an eighth lens (L8) having negative refractive power. An image-side surface (S14) of the seventh lens (L7) is a concave surface at the optical axis. In addition, the optical system (100) satisfies 1<TTL/<2.5, wherein TTL is the distance between the object-side surface (S1) of the first lens (L1) and an imaging surface (S19) of the optical system (100) on the optical axis. The optical system (100) further comprises a diaphragm (STO), and L is the effective aperture diameter of the diaphragm (STO).

Abstract: A magnetic resonance imaging method includes: obtaining three-dimensional under-sampling data of a target object based on a first three-dimensional magnetic resonance imaging sequence; obtaining a three-dimensional point spread function based on the three-dimensional under-sampling data or a two-dimensional mapping data of the target object; obtaining a sensitivity map of the target object based on the data collected by three-dimensional low-resolution complete sampling; performing imaging reconstruction to the three-dimensional under-sampling data based on the three-dimensional point spread function and the sensitivity map to obtain a reconstructed magnetic resonance image. The first three-dimensional magnetic resonance imaging sequence has a first sinusoidal gradient field on a phase direction and a second sinusoidal gradient field on a layer selection direction. 0-order moments of the first and the second three-dimensional magnetic resonance imaging sequences are 0.

Abstract: A device for sampling soil of tropical lowland rainforest is provided, and the device includes a driving motor, a probe rod and a sampling barrel. An output shaft of the driving motor is fixedly connected with the top end of the probe rod. The probe rod is hollow. The sampling barrel is disposed in the probe rod. An opening of the bottom end of the sampling barrel faces to the soil inlet. Two baffle plates are also arranged in the probe rod. The side wall of the baffle plate is arc-shaped, and the bottom end of the baffle plate is semicircular. The semicircular baffle plates can pass through the space and can be joined to form a cylinder with a closed bottom end. Two linear motors are fixedly arranged in the probe rod and can respectively drive the two baffle plates to move.

Abstract: An apriori guidance network for multitask medical image synthesis is provided. The apriori guidance network includes a generator and a discriminator, wherein the generator includes an apriori guidance module configured to convert an input feature map into a target modal image pointing to a target domain according to an apriori feature, and the apriori feature is a deep feature of the target modal image. The generator is configured to generate a corresponding target domain image by taking the apriori feature of the target modal image and source modal image data as an input. The discriminator is configured to discriminate an authenticity of the target domain image outputted by the generator.

Abstract: Neural Architecture Search (NAS) is a laborious process. Prior work on automated NAS targets mainly on improving accuracy but lacked consideration of computational resource use. Presented herein are embodiments of a Resource-Efficient Neural Architect (RENA), an efficient resource-constrained NAS using reinforcement learning with network embedding. RENA embodiments use a policy network to process the network embeddings to generate new configurations. Example demonstrates of RENA embodiments on image recognition and keyword spotting (KWS) problems are also presented herein. RENA embodiments can find novel architectures that achieve high performance even with tight resource constraints. For the CIFAR10 dataset, the tested embodiment achieved 2.95% test error when compute intensity is greater than 100 FLOPs/byte, and 3.87% test error when model size was less than 3M parameters.

Abstract: Presented herein are embodiments of a prefix-to-prefix framework for simultaneous translation that implicitly learns to anticipates in a single translation. Within these frameworks are effective “wait-k” policy model embodiments that may be trained to generate a target sentence concurrently with a source sentence but lag behind by a predefined number of words. Embodiments of the prefix-to-prefix framework achieve low latency and better quality when compared to full-sentence translation in four directions: Chinese?English and German?English. Also presented herein is a novel latency metric that addresses deficiencies of previous latency metrics.

Abstract: Described herein are systems and methods to identify and address sources of bias in an end-to-end speech model. In one or more embodiments, the end-to-end model may be a recurrent neural network with two 2D-convolutional input layers, followed by multiple bidirectional recurrent layers and one fully connected layer before a softmax layer. In one or more embodiments, the network is trained end-to-end using the CTC loss function to directly predict sequences of characters from log spectrograms of audio. With optimized recurrent layers and training together with alignment information, some unwanted bias induced by using purely forward only recurrences may be removed in a deployed model.

Abstract: Presented herein are embodiments of a prefix-to-prefix framework for simultaneous translation that implicitly learns to anticipates in a single translation. Within these frameworks are effective “wait-k” policy model embodiments that may be trained to generate a target sentence concurrently with a source sentence but lag behind by a predefined number of words. Embodiments of the prefix-to-prefix framework achieve low latency and better quality when compared to full-sentence translation in four directions: Chinese?English and German?English. Also presented herein is a novel latency metric that addresses deficiencies of previous latency metrics.

Abstract: Neural Architecture Search (NAS) is a laborious process. Prior work on automated NAS targets mainly on improving accuracy but lacked consideration of computational resource use. Presented herein are embodiments of a Resource-Efficient Neural Architect (RENA), an efficient resource-constrained NAS using reinforcement learning with network embedding. RENA embodiments use a policy network to process the network embeddings to generate new configurations. Example demonstrates of RENA embodiments on image recognition and keyword spotting (KWS) problems are also presented herein. RENA embodiments can find novel architectures that achieve high performance even with tight resource constraints. For the CIFAR10 dataset, the tested embodiment achieved 2.95% test error when compute intensity is greater than 100 FLOPs/byte, and 3.87% test error when model size was less than 3M parameters.

Abstract: Described herein are systems and methods for automatic unit selection and target decomposition for sequence labelling. Embodiments include a new loss function called Gram-Connectionist Temporal Classification (CTC) loss that extend the popular CTC loss function criterion to alleviate prior limitations. While preserving the advantages of CTC, Gram-CTC automatically learns the best set of basic units (grams), as well as the most suitable decomposition of target sequences. Unlike CTC, embodiments of Gram-CTC allow a model to output variable number of characters at each time step, which enables the model to capture longer term dependency and improves the computational efficiency. It is also demonstrated that embodiments of Gram-CTC improve CTC in terms of both performance and efficiency on the large vocabulary speech recognition task at multiple scales of data, and that systems that employ an embodiment of Gram-CTC can outperform the state-of-the-art on a standard speech benchmark.

Abstract: Described herein are systems and methods to identify and address sources of bias in an end-to-end speech model. In one or more embodiments, the end-to-end model may be a recurrent neural network with two 2D-convolutional input layers, followed by multiple bidirectional recurrent layers and one fully connected layer before a softmax layer. In one or more embodiments, the network is trained end-to-end using the CTC loss function to directly predict sequences of characters from log spectrograms of audio. With optimized recurrent layers and training together with alignment information, some unwanted bias induced by using purely forward only recurrences may be removed in a deployed model.

Abstract: Described herein are systems and methods for automatic unit selection and target decomposition for sequence labelling. Embodiments include a new loss function called Gram-Connectionist Temporal Classification (CTC) loss that extend the popular CTC loss function criterion to alleviate prior limitations. While preserving the advantages of CTC, Gram-CTC automatically learns the best set of basic units (grams), as well as the most suitable decomposition of target sequences. Unlike CTC, embodiments of Gram-CTC allow a model to output variable number of characters at each time step, which enables the model to capture longer term dependency and improves the computational efficiency. It is also demonstrated that embodiments of Gram-CTC improve CTC in terms of both performance and efficiency on the large vocabulary speech recognition task at multiple scales of data, and that systems that employ an embodiment of Gram-CTC can outperform the state-of-the-art on a standard speech benchmark.

Abstract: A system and method for 3D design includes defining a three-dimensional virtual interaction space visualized with a 3D camera system operable to generate three-dimensional coordinate data corresponding to physical objects within the interaction space. A physical gesture of the user within the interaction space is interpreted according to pre-determined rules, the physical gesture including a movement of a physical object. A virtual shape is generated or manipulated in response to the interpretation of the physical gesture, the virtual shape residing virtually within the interaction space. A representation of the virtual 3D shape is interactively displayed during the physical gesture.