Alibaba Group Holding Limited

Acbel Polytech Inc.

Tao Zhang

Ning Guo

Jiaqing Guo

Pengcheng Li

Yongbin Gu

Xiao-Lin Ma

The present disclosure provides methods, systems, and non-transitory computer readable media for fetching data for an accelerator. The methods include detecting an attempt to access a first page of data that is not stored on a primary storage unit of the accelerator, wherein the first page of data corresponds to part of an array with a number of dimensions; and responsive to detecting the attempt to access the first page of data: dividing the array into sub-arrays by: dividing the array into a number of 1st level sub-arrays, and dividing a first 1st level sub-array into a number of 2nd level sub-arrays, wherein the first 1st level sub-array contains the first page of data; selecting pages for prefetching, wherein selecting pages for prefetching includes selecting all pages in a first 2nd level sub-array for prefetching if the first 2nd level sub-array meets a page access volume condition, wherein the first 2nd level sub-array contains the first page of data; and transferring the first page of data and any pages of data selected for prefetching from a memory system connected to the accelerator into the primary storage unit.

Methods and systems for fetching data for an accelerator

The present disclosure provides methods, systems, and non-transitory computer readable media for fetching data for an accelerator. The methods include detecting an attempt to access a first page of data that is not stored on a primary storage unit of the accelerator; and responsive to detecting the attempt to access the first page of data: assessing activity of the accelerator; determining, based on the assessed activity of the accelerator, a prefetch granularity size; and transferring a chunk of contiguous pages of data of the prefetch granularity size from a memory system connected to the accelerator into the primary storage unit, wherein the transferred chunk of contiguous pages of data include the first page of data.

Techniques for returning estimated value of search keywords of an entire account include, for the entire account, obtaining one or more selected search keywords and their respective forecast periods and parameter settings. An estimated value of a respective search keyword in the respective forecast period is forecasted. Based on stored historical data and parameter settings of the respective search keyword, the estimated value of the respective search keyword is modified to obtain a modified estimated value. The modified estimated value of each search keyword is added up to generate an estimated value of the entire account. The estimated value of the entire account is returned to a client terminal from which the entire account is logged in. The present disclosure modifies the respective search keyword's estimated value so that the estimated value of the entire account satisfies the expected value of the client.

Returning estimated value of search keywords of entire account

The present disclosure introduces a method, an apparatus and memory of predicting a user behavior number of a word for reducing the amount and the complexity of operation, saving the consumption of the equipment, and improving the accuracy and reliability of predictions. In an embodiment, a historical data sequence of the user behavior number of a word is converted from a time domain to a frequency domain. Based on the converted frequency domain, each estimated cycle and its effect rate value of the historical data sequence are ascertained. If the historical data sequence is stable, an average value of user behavior numbers of some historical data points before a prediction point is calculated as a user behavior number of the prediction point. Otherwise, the user behavior number is calculated based on a selected main cycle and a selected singularity.

Predicting a user behavior number of a word

Providing online advertisements includes selecting, among a plurality of advertising sources provided by a seller, a selected plurality of advertising sources that meet a predefined condition; generating a plurality of keyword sets that correspond to the selected plurality of advertising sources; establishing a programming model according to a set of predefined constraints, wherein the programming model represents match of the selected plurality of advertising sources and the plurality of keyword sets; and determining a substantially optimal match between at least some of the plurality of advertising sources and the plurality of keyword sets by solving the programming model.

Matching of advertising sources and keyword sets in online commerce platforms

A power supply for a negative voltage load has a switch-mode power unit, a monitoring unit, a first auxiliary power unit and a second auxiliary power unit. The switch-mode power unit has a rectifying circuit, a power factor correction circuit, a DC to DC power circuit. The first auxiliary power unit is connected to output terminals of the power factor correction circuit and converts DC power to a first DC power for driving the monitoring unit. The second auxiliary power unit is connected to the negative voltage load in parallel and converts negative power from the negative voltage load to a second DC power for driving the monitoring unit when AC power is interrupted. Since the first and the second auxiliary power unit do not work at the same time, the power supply can effectively enhances the conversion efficiency of the power supply.

Growing community of inventors

Hangzhou, China

Tao Zhang