Adobe, Inc.

Deepak Pai

Shankar Venkitachalam

Debraj Basu

Joshua Sweetkind-Singer

Vijay Srivastava

Nikaash Puri

Vinh Ngoc Khuc

Nimish Srivastav

Balaji Krishnamurthy

Trung T Nguyen

Mihir Naware

Abhishek Pani

Yaman Kumar

Kumar Mrityunjay Singh

Hyman Chung

Kunal Kumar Jain

Simra Shahid

Sukriti Verma

Meghanath M Y

Florin Silviu Iordache

Sandipan Karmakar

Sy Bor Wang

Sean Steimer

Horea Bacila

Umang Moorarka

Neha Gupta

Shirsh Bansal

Jose Mathew

Shradha Agrawal

Niyati Himanshu Chhaya

Atanu Ranjan Sinha

Ritwik Sinha

Kushal Chawla

Dhruv Singal

Ganesh Satish Mallya

Craig Mathis

Anil Kamath

Ponnurangam Kumaraguru

James Harold Brown

Joshua Aaron Hansen

Sriram Ravindran

Paridhi Jain

Yash Shrivastava

Ambareesh Revanur

Dhwanit Agarwal

Anish Narang

Priyam Tejaswin

Kenneth Hahn

Shubhranshu Shekhar

Dhanya Raghu

Some aspects relate to technologies for generating time series of sketches and using the time series of sketches for approximate query processing. In accordance with some aspects, tabular data is accessed that has a number of columns. Responsive to identifying a first column as comprising numerical data, sketches are generated for the numerical data for each of a number of time steps, and the sketches for the numerical data are stored as a first time series of sketches. Responsive to identifying a second column as comprising categorical data, sketches are generated for the categorical data for each of the time steps, and the sketches for the categorical data are stored as a second time series of sketches. When a query is received, a response to the query is provided using sketches from the first time series of sketches and/or the second time series of sketches.

Time series of sketches for approximate query processing

In accordance with the described techniques, an image transformation system receives an input image and a text prompt, and leverages a generator network to edit the input image based on the text prompt. The generator network includes a plurality of layers configured to perform respective edits. A plurality of masks are generated based on the text prompt that define local edit regions, respectively, of the input image for respective layers of the generator network. Further, the generator network generates an edited image by editing the input image based on the plurality of masks, the respective edits of the respective layers, and the text prompt.

Mask conditioned image transformation based on a text prompt

Methods, computer systems, computer-storage media, and graphical user interfaces are provided for determining user affinities by tracking historical user interactions with tagged digital content and using the user affinities in content generation applications. Accordingly, the system may track user interactions with published digital content in order to generate user interaction reports whenever a user engages with the digital content. The system may aggregate the interaction reports to generate an affinity profile for a user or audience of users. A marketer may then generate digital content for a target user or audience of users and the system may process the digital content to generate a set of tags for the digital content. Based on the set of tags, the system may then evaluate the digital content in view of the affinity profile for the target user/audience to determine similarities or differences between the digital content and the affinity profile.

Determining user affinities for content generation applications

Enhanced methods for improving the performance of classifiers are described. A ground-truth labeled dataset is accessed. A classifier predicts a predicted label for datapoints of the dataset. A confusion matrix for the dataset and classifier is generated. A credibility interval is determined for a performance metric for each label. A first labels with a sufficiently large credibility interval is identified. A second label is identified, where the classifier is likely to confuse, in its predictions, the first label with the second label. The identification of the second label is based on instances of incorrect label predictions of the classifier for the first and/or the second labels. The classifier is updated based on a new third label that includes an aggregation of the first label and the second label. The updated classifier model predicts the third label for any datapoint that the classifier previously predicted the first or second labels.

Quantifying and improving the performance of computation-based classifiers

The present disclosure relates to systems, methods, and non-transitory computer readable media that utilize intelligent contextual bias weights for informing keyphrase relevance models to extract keyphrases. For example, the disclosed systems generate a graph from a digital document by mapping words from the digital document to nodes of the graph. In addition, the disclosed systems determine named entity bias weights for the nodes of the graph utilizing frequencies with which the words corresponding to the nodes appear within named entities identified from the digital document. Moreover, the disclosed systems generate a keyphrase summary for the digital document utilizing the graph and a machine learning model biased according to the named entity bias weights for the nodes of the graph.

Generating weighted contextual themes to guide unsupervised keyphrase relevance models

In some embodiments, a computing system identifies a current engagement stage of a user with an online platform by applying a stage prediction model based on interaction data associated with the user. The interaction data describe actions performed by the user with respect to the online platform and context data associated with each of the actions. The computing system further identifies one or more critical events for promoting the user to transition from one engagement stage to a higher engagement stage based on the interaction data associated with the user. The computing system can make the identified current engagement stage of the user or the identified critical event to be accessible by the online platform so that user interfaces presented on the online platform can be modified to improve a likelihood of the user to transit from the current stage to a higher engagement stage.

Machine learning models applied to interaction data for facilitating modifications to online environments

In some examples, a prototype model that includes a representative subset of data points (e.g., inputs and output classifications) of a machine learning model is analyzed to efficiently interpret the machine learning model's behavior. Performance metrics such as a critic fraction, local explanation scores, and global explanation scores are determined. A local explanation score capture an importance of a feature of a test point to the machine learning model determining a particular class for the test point and is computed by comparing a value of a feature of a test point to values for prototypes of the prototype model. Using a similar approach, global explanation scores may be computed for features by combining local explanation scores for data points. A critic fraction may be computed to quantify a misclassification rate of the prototype model, indicating the interpretability of the model.

Prototype-based machine learning reasoning interpretation

Methods, systems, and non-transitory computer readable storage media are disclosed for analyzing feature impact of a machine-learning model using prototypes across analytical spaces. For example, the disclosed system can identify features of data points used to generate outputs via a machine-learning model and then map the features to a feature space and the outputs to a label space. The disclosed system can then utilize an iterative process to determine prototypes from the data points based on distances between the data points in the feature space and the label space. Furthermore, the disclosed system can then use the prototypes to determine the impact of the features within the machine-learning model based on locally sensitive directions; region variability; or mean, range, and variance of features of the prototypes.

Determining feature impact within machine learning models using prototypes across analytical spaces

In various embodiments of the present disclosure, output data generated by a deployed machine learning model may be received. An input data anomaly may be detected based at least in part on analyzing input data of the deployed machine learning model. An output data anomaly may further be detected based at least in part on analyzing the output data of the deployed machine learning model. A determination may be made that the input data anomaly contributed to the output data anomaly based at least in part on comparing the input data anomaly to the output data anomaly. A report may be generated that is indicative of the input data anomaly and the output data anomaly, and the report may be transmitted to a client device.

Anomaly detection and reporting for machine learning models

The present disclosure relates to utilizing a graph convolutional neural network to generate similarity probabilities between pairs of digital identities associated with digital transactions based on time dependencies for use in identifying fraudulent transactions. For example, the disclosed systems can generate a transaction graph that includes nodes corresponding to digital identities. The disclosed systems can utilize a time-dependent graph convolutional neural network to generate node embeddings for the nodes based on the edge connections of the transaction graph. Further, the disclosed systems can utilize the node embeddings to determine whether a digital identity is associated with a fraudulent transaction.

Utilizing a time-dependent graph convolutional neural network for fraudulent transaction identification

Embodiments of the present disclosure are directed towards generating images conditioned on a desired attribute. In particular, an attribute-based image generation system can use a directional-GAN architecture to generate images conditioned on a desired attribute. A latent vector and a desired attribute are received. A feature subspace is determined for the latent vector using a latent-attribute linear classifier trained to determine a relationship between the latent vector and the desired attribute. An image is generated using the latent vector such that the image contains the desired attribute. In embodiments, where the feature space differs from a desired feature subspace, a directional vector is applied to the latent vector that shifts the latent vector from the feature subspace to the desired feature subspace. This modified latent vector is then used during generation of the image.

Generation of controlled attribute-based images

Forecasting a potential audience size and an unduplicated audience size for a digital campaign includes receiving an audience segment input and a time period input. The audience segment input is converted into multiple atomic target specifications. For each of the multiple atomic target specifications, a potential audience size is determined during the time period input by selecting a time series model based on a frequency of attribute values from the atomic target specification and combining the selected time series model with a frequent item set model. The potential audience size for each of atomic target specifications is aggregated over the time period input into a total potential audience size. The total potential audience size is output. The time series model and the frequent item set model are obtained using data from a historic bid request database.

Forecasting potential audience size and unduplicated audience size

Systems and methods are disclosed herein for distributing online ads with electronic content according to online ad request targeting parameters. One embodiment of this technique involves placing online test ads across multiple online ad request dimensions and tracking a performance metric for the online test ads. The performance of the online ad request dimensions is estimated based on the tracking of the performance metric for the online test ads and online ad request targeting parameters are established for spending a budget of a campaign to place online ads in response to online ad requests having particular online ad request dimensions. Online ads are then distributed based on using the online ad request targeting parameters to select online ad requests.

Methods for determining targeting parameters and bids for online ad distribution

Growing community of inventors

Santa Clara, CA, United States of America

Deepak Pai