Showing papers by "Sameep Mehta published in 2019"

FactSheets: Increasing trust in AI services through supplier's declarations of conformity

Abstract: Accuracy is an important concern for suppliers of artificial intelligence (AI) services, but considerations beyond accuracy, such as safety (which includes fairness and explainability), security, and provenance, are also critical elements to engender consumers’ trust in a service. Many industries use transparent, standardized, but often not legally required documents called supplier's declarations of conformity (SDoCs) to describe the lineage of a product along with the safety and performance testing it has undergone. SDoCs may be considered multidimensional fact sheets that capture and quantify various aspects of the product and its development to make it worthy of consumers’ trust. In this article, inspired by this practice, we propose FactSheets to help increase trust in AI services. We envision such documents to contain purpose, performance, safety, security, and provenance information to be completed by AI service providers for examination by consumers. We suggest a comprehensive set of declaration items tailored to AI in the Appendix of this article.

Think Your Artificial Intelligence Software Is Fair? Think Again

Abstract: Today, machine-learning software is used to help make decisions that affect people's lives. Some people believe that the application of such software results in fairer decisions because, unlike humans, machine-learning software generates models that are not biased. Think again. Machine-learning software is also biased, sometimes in similar ways to humans, often in different ways. While fair model- assisted decision making involves more than the application of unbiased models-consideration of application context, specifics of the decisions being made, resolution of conflicting stakeholder viewpoints, and so forth-mitigating bias from machine-learning software is important and possible but difficult and too often ignored.

Ownership Preserving AI Market Places Using Blockchain

Abstract: We present a blockchain based system that allows data owners, cloud vendors, and AI developers to collaboratively train machine learning models in a trustless AI marketplace. Data is a highly valued digital asset and central to deriving business insights. Our system enables data owners to retain ownership and privacy of their data, while still allowing AI developers to leverage the data for training. Similarly, AI developers can utilize compute resources from cloud vendors without loosing ownership or privacy of their trained models. Our system protocols are set up to incentivize all three entities - data owners, cloud vendors, and AI developers to truthfully record their actions on the distributed ledger, so that the blockchain system provides verifiable evidence of wrongdoing and dispute resolution. Our system is implemented on the Hyperledger Fabric and can provide a viable alternative to centralized AI systems that do not guarantee data or model privacy. We present experimental performance results that demonstrate the latency and throughput of its transactions under different network configurations where peers on the blockchain may be spread across different datacenters and geographies. Our results indicate that the proposed solution scales well to large number of data and model owners and can train up to 70 models per second on a 12-peer non optimized blockchain network and roughly 30 models per second in a 24 peer network.

A novel feature transform framework using deep neural network for multimodal floor plan retrieval

Abstract: In recent past, there has been a steep increase in the use of online platforms for the search of desired products. Real estate industry is no exception and has started initiating rent/sale of houses through online platforms. In this paper, we propose a deep neural network framework to facilitate automatic search of homes based on their floor plans. The salient features of this framework are that the query can be either an image (existing floor plan) or a sketch through a sketch pad interface. Our proposed framework automatically determines the type of query (image or sketch) and retrieves similar floor plan images from the database. The critical contributions of our proposed approach are: (1) a novel unified floor plan retrieval framework using multimodal query, i.e., an intuitive and convenient sketch query mode as well as a query by example mode ; (2) a conjunction of autoencoder, Cyclic GAN and CNN for the task of domain mapping and floor plan image retrieval. We have reported results of extensive experimentation and comparison with baseline results to establish the effectiveness of our approach.

Hardening Deep Neural Networks via Adversarial Model Cascades

Abstract: Deep neural networks (DNNs) are vulnerable to malicious inputs crafted by an adversary to produce erroneous outputs. Works on securing neural networks against adversarial examples achieve high empirical robustness on simple datasets such as MNIST. However, these techniques are inadequate when empirically tested on complex data sets such as CIFAR-10 and SVHN. Further, existing techniques are designed to target specific attacks and fail to generalize across attacks. We propose Adversarial Model Cascades (AMC) as a way to tackle the above inadequacies. Our approach trains a cascade of models sequentially where each model is optimized to be robust towards a mixture of multiple attacks. Ultimately, it yields a single model which is secure against a wide range of attacks; namely FGSM, Elastic, Virtual Adversarial Perturbations and Madry. On an average, AMC increases the model’s empirical robustness against various attacks simultaneously, by a significant margin (of 6.225% for MNIST, 5.075% for SVHN and 2.65% for CIFAR-10 ). At the same time, the model’s performance on non-adversarial inputs is comparable to the state-of-the-art models.

On Efficiently Processing Workflow Provenance Queries in Spark

Abstract: In this paper, we look at how we can leverage Spark platform for efficiently processing fine-grained provenance queries on large volumes of workflow provenance data. Simple recursive querying based Spark solutions involve large data scanning cost and hence do not work well. We propose a novel provenance framework which is engineered to quickly determine a small volume of data containing the entire lineage of the queried data-item. This small volume of data is then recursively processed to figure out the provenance of the queried data-item. We study the effectiveness of the proposed framework on a provenance trace obtained from a financial domain text curation workflow and report our observations. We show that the proposed framework easily outperforms the naive approaches.

Schema-free in-graph indexing

Abstract: Methods, systems, and computer program products for schema-free in-graph indexing are provided herein. A computer-implemented method includes creating multiple indexes directed to data within a knowledge graph; correlating two or more of the created indexes, thereby generating one or more multi-dimensional indexes; determining, based on a received query, one or more traversal paths within the data of the knowledge graph and the generated multi-dimensional indexes, wherein the traversal paths facilitate processing of the query; and outputting a response to the query based on the determined traversal paths.

Radial Loss for Learning Fine-grained Video Similarity Metric

Abstract: In this paper, we propose the Radial Loss which utilizes category and sub-category labels to learn an order-preserving fine-grained video similarity metric. We propose an end-to-end quadlet-based Convolutional Neural Network (CNN) combined with Long Short-term Memory (LSTM) Unit to model video similarities by learning the pairwise distance relationships between samples in a quadlet generated using the category and sub-category labels. We showcase two novel applications of learning a video similarity metric - (i) fine-grained video retrieval, (ii) fine-grained event detection, along with simultaneous shot boundary detection, and correspondingly show promising results against those of the baselines on two new fine-grained video datasets.

Half-pyramid data encryption

Abstract: One embodiment provides a method, including: receiving, from a data owner, an input string of plaintext data comprising a plurality of characters for storage in a database of a third-party storage provider; arranging the plurality of characters of the input string as a half pyramid, wherein the half pyramid comprises a plurality of rows, each row comprising at least one more character than a preceding row; encrypting, using a secure encryption scheme and based upon a key, each row of the half pyramid independently from each other row of the half pyramid; and storing, in the database of the third-party storage provider, the encrypted rows of the half pyramid. Other aspects are claimed and described.

Fair Transfer of Multiple Style Attributes in Text

Abstract: To preserve anonymity and obfuscate their identity on online platforms users may morph their text and portray themselves as a different gender or demographic. Similarly, a chatbot may need to customize its communication style to improve engagement with its audience. This manner of changing the style of written text has gained significant attention in recent years. Yet these past research works largely cater to the transfer of single style attributes. The disadvantage of focusing on a single style alone is that this often results in target text where other existing style attributes behave unpredictably or are unfairly dominated by the new style. To counteract this behavior, it would be nice to have a style transfer mechanism that can transfer or control multiple styles simultaneously and fairly. Through such an approach, one could obtain obfuscated or written text incorporated with a desired degree of multiple soft styles such as female-quality, politeness, or formalness. To the best of our knowledge this work is the first that shows and attempt to solve the issues related to multiple style transfer. We also demonstrate that the transfer of multiple styles cannot be achieved by sequentially performing multiple single-style transfers. This is because each single style-transfer step often reverses or dominates over the style incorporated by a previous transfer step. We then propose a neural network architecture for fairly transferring multiple style attributes in a given text. We test our architecture on the Yelp dataset to demonstrate our superior performance as compared to existing one-style transfer steps performed in a sequence.

Stolen machine learning model identification

Abstract: One embodiment provides a method, including: assigning a machine learning model signature to a machine learning model, wherein the machine learning model signature is generated using (i) data points and (ii) corresponding data labels from training data; receiving input comprising identification of a target machine learning model; acquiring a target signature for the target machine learning model by generating a signature for the target machine learning model using (i) data points from the assigned machine learning model signature and (ii) labels assigned to those data points by the target machine learning model; determining a stolen score by comparing the target signature to the machine learning model signature and identifying the number of data labels that match between the target signature and the machine learning model signature; and classifying the target machine learning model as stolen based upon the stolen score reaching a predetermined threshold.

Handling Gender Biases in E-Commerce Product Specifications

Abstract: Fair computing has emerged as a key area of artificial intelligence (AI), and especially machine learning (ML). Identification and mitigation of several types of biases, spanning over data and machine learning models, has attracted both research and regulatory attention. In this work, we explore the presence and degree of gender bias in product descriptions featured on e-commerce websites. Using the knowledge obtained in analysis, we recommend methods to debias the product description, using a product feature level text selection scheme, sourced by customer reviews. Our work is the first of its kind, that establishes a baseline for enhancing the cross-gender acceptability of product descriptions, and proposes a framework for e-retailers to provide such gender-neutral product descriptions.

Learning Convolutional Neural Networks with Deep Part Embeddings

Abstract: We propose a novel concept of Deep Part Embeddings (DPEs), which can be used to learn new Convolutional Neural Networks (CNNs) for different classes. We define DPE as a neuron of a trained CNN along with its network of filter activations that is interpretable as a part of a class that the neuron contributes to. Given a new class $\mathcal{C}$, we explore the idea of combining different DPEs that intuitively constitute $\mathcal{C}$, from trained CNNs (not on $\mathcal{C}$), into a network that learns the class $\mathcal{C}$ with few training samples. An important application of our proposed framework is the ability to modify a CNN trained on n classes to learn a new class with limited training data without significantly affecting its performance on the n classes. We visually illustrate the different network architectures and extensively evaluate their performance against the baselines.

Generating a recommended shaping function to integrate data within a data repository

Abstract: A method includes determining, by a controller, a portion of data that is selected by a user. The portion of data includes source data that is to be transformed by at least one shaping function. The method also includes generating, by the controller, a first output recommendation data that communicates at least one recommended shaping function to apply to the portion of data. The first output recommendation data is generated based on patterns of shaping functions that have been previously chosen. The patterns of shaping functions that have been previously chosen can be chosen by a plurality of system users. The method also includes determining whether to apply the at least one recommended shaping function to the portion of data. The method also includes applying the at least one recommended shaping function based on the determining.

Design diagrams as ontological source

Abstract: begin In custom software development projects, it is frequently the case that the same type of software is being built for different customers. The deliverables are similar because they address the same market (e.g., Telecom, Banking) or have similar functions or both. However, most organisations do not take advantage of this similarity and conduct each project from scratch leading to lesser margins and lower quality. Our key observation is that the similarity among the projects alludes to the existence of a veritable domain of discourse whose ontology, if created, would make the similarity across the projects explicit. Design diagrams are an integral part of any commercial software project deliverables as they document crucial facets of the software solution. We propose an approach to extract ontological information from UML design diagrams (class and sequence diagrams) and represent it as domain ontology in a convenient representation. This ontology not only helps in developing a better understanding of the domain but also fosters software reuse for future software projects in that domain. Initial results on extracting ontology from thousands of model from public repository show that the created ontologies are accurate and help in better software reuse for new solutions. endabstract