Abstract:
A neural network accelerator architecture for multiple task adaptation comprises a volatile memory comprising a plurality of subarrays, each subarray comprising M rows and N columns of volatile memory cells; a source line driver connected to a plurality of N source lines, each source line corresponding to a column in the subarray; a binary mask buffer memory having size at least N bits, each bit corresponding to a column in the subarray, where a 0 corresponds to turning off the column for a convolution operation and a 1 corresponds to turning on the column for the convolution operation; and a controller configured to selectively drive each of the N source lines with a corresponding value from the mask buffer; wherein each column in the subarray is configured to store a convolution kernel.
Abstract:
A terminal block structure (1, 1A) is provided and includes an insulative housing (10), separation members (20, 20A) and conductive wire and terminal crimping members (30). The separation member (20, 20A) is secured to the insulative housing and includes a base (21) and a pressing mechanism (22) operably connected to the base (21). The pressing mechanism (22) includes a pressing portion (221) and a tail portion (222) extended outward from the pressing portion (221). The conductive wire and terminal crimping members (30) are arranged inside the insulative housing (10). When an acting force is exerted on the pressing portion (221), the pressing mechanism (22) generates deformation or rotation to drive the tail portion (222) to ascend. Accordingly, the pressing operation of the terminal block structure (1, 1A) is facilitated, and it is equipped with simple structure for easy installation.
Abstract:
Dynamic additive attention adaption for memory-efficient multi-domain on-device learning is provided. Almost all conventional methods for multi-domain learning in deep neural networks (DNNs) only focus on improving accuracy with minimal parameter update, while ignoring high computing and memory cost during training. This makes it difficult to deploy multi-domain learning into resource-limited edge devices, like mobile phones, internet-of-things (IoT) devices, embedded systems, and so on. To reduce training memory usage, while keeping the domain adaption accuracy performance, Dynamic Additive Attention Adaption (DA3) is proposed as a novel memory-efficient on-device multi-domain learning approach. Embodiments of DA3 learn a novel additive attention adaptor module, while freezing the weights of the pre-trained backbone model for each domain. This module not only mitigates activation memory buffering for reducing memory usage during training, but also serves as a dynamic gating mechanism to reduce the computation cost for fast inference.
Abstract:
A terminal block structure (1, 1A) is provided and includes an insulative housing (10), separation members (20, 20A) and conductive wire and terminal crimping members (30). The separation member (20, 20A) is secured to the insulative housing and includes a base (21) and a pressing mechanism (22) operably connected to the base (21). The pressing mechanism (22) includes a pressing portion (221) and a tail portion (222) extended outward from the pressing portion (221). The conductive wire and terminal crimping members (30) are arranged inside the insulative housing (10). When an acting force is exerted on the pressing portion (221), the pressing mechanism (22) generates deformation or rotation to drive the tail portion (222) to ascend. Accordingly, the pressing operation of the terminal block structure (1, 1A) is facilitated, and it is equipped with simple structure for easy installation.
Abstract:
The invention relates to a terminal block with a handle. The terminal block includes an insulative base, a conductive terminal, a flexible clamp and a toggle member. The insulative base has a chamber and an insertion hole communicating with the chamber. The conductive terminal is fixed at a bottom of the insertion hole. The flexible clamp is received in the chamber and extended with a movable elastic arm abutting against the conductive terminal and closing the inserting hole. The toggle member has a pivot rotatably connected in the chamber. The pivot is extended with a pusher abutting against the movable elastic arm. When the toggle member is rotated to a release position, the pusher is rotated about the pivot to press the movable elastic arm so the movable elastic arm deviates from the conductive terminal and the insertion hole is opened.
Abstract:
An electrolyte solvent for a cathode active material composed of lithium oxo acid salt. The solvent is used for a lithium ion secondary battery using the lithium oxo acid salt as a cathode material. The electrolyte solvent includes an ammonium ion which includes two or more alkoxyalkyl groups.
Abstract:
A microfluidic electrochemical device and process are detailed that provide chemical imaging and electrochemical analysis under vacuum at the surface of the electrode-sample or electrode-liquid interface in-situ. The electrochemical device allows investigation of various surface layers including diffuse layers at selected depths populated with, e.g., adsorbed molecules in which chemical transformation in electrolyte solutions occurs.
Abstract:
In an embodiment, a method for manufacturing a thin layer chromatography (“TLC”) plate is disclosed. The method includes forming a layer of elongated nanostructures (e.g., carbon nanotubes), and at least partially coating the elongated nanostructures with a coating. The coating includes a stationary phase and/or precursor of a stationary phase for use in chromatography. Embodiments for TLC plates and related methods are also disclosed.
Abstract:
Methods and apparatus relate to the synthesis of high fidelity polynucleotides and to the reduction of sequence errors generated during synthesis of nucleic acids on a solid support. Specifically, design of support-bound template oligonucleotides is disclosed. Assembly methods include cycles of annealing, stringent wash and extension of polynucleotides comprising a sequence region complementary to immobilized template oligonucleotides. The error free synthetic nucleic acids generated therefrom can be used for a variety of applications, including synthesis of biofuels and value-added pharmaceutical products.