International Business Machines Corporation

Doordash, Inc.

Biplav Srivastava

Kartik Talamadupula

Srikanth Govindaraj Tamilselvam

Richard Thomas Goodwin

Rakesh Mohan

Debdoot Mukherjee

Vera Liao

Juhnyoung Lee

Vibha Singhal Sinha

Vijay Krishnarao Naik

Pietro Mazzoleni

Yunfeng Zhang

Maja Vukovic

Amit Anil Nanavati

Jim Alain Laredo

Krishna Kummamuru

Elizabeth Daly

Oznur Alkan

Vinod Muthusamy

Aleksander A Slominski

Sumit Mittal

Shirin Sohrabi Araghi

Rema Ananthanarayanan

Murray Scott Campbell

Qingzi Vera Liao

Matthew Scott Davis

Ullas Balan Nambiar

Aubrey J Rembert

Pavan Kapanipathi Bangalore

Prithu Banerjee

Anca Andreea Ivan

Raj Gupta

Maharshi H Desai

Lydia Manikonda

Kuntal Dey

Arun Kumar

Anton Viktorovich Riabov

Deepak Srinivas Turaga

Heiko Hary Ludwig

Adi Ionel Botea

Kalapriya Kannan

Miao Liu

Akihiro Kishimoto

Pankaj Dhoolia

Anuradha Bhamidipaty

Nathalie Baracaldo Angel

Radu Marinescu

Guruduth Somasekhara Banavar

Sudhanshu Shekhar Singh

Koustuv Dasgupta

Girish Bhimrao Chafle

Terrence E White

Vivek Tyagi

Bryant R Chen

Ankur Jain

Rakesh Rameshrao Pimplikar

Sweefen Goh

Senthil K Mani

Ramana V Polavarapu

Manisha Dattatraya Bhandar

Francesca Rossi

Manish P Kurhekar

Sumana S Bhat

Karen L B Brown

Bipin K Joshi

Pramod Anantharam

Ashish Kundu

Manish Anand Bhide

Xin Zhou

Florian Pinel

Krishna Chaitanya Ratakonda

Manoj N Kumar

Nitendra Rajput

Shivkumar Kalyanaraman

Jie Cui

Yuan-Chi Chang

Qian Ma

Prasad Manikarao Deshpande

Vikas Agarwal

Aleksandra Mojsilovic

Rachel K E Bellamy

Shyh-Kwei Chen

John Erik Wittern

Manikandan Padmanaban

Daniel V Oppenheim

Nanjangud C Narendra

Natwar Modani

Deepak S Padmanabhan

Parul Alok Mittal

Vishal Singh Batra

Kshitij P Fadnis

Karthikeyan Ponnalagu

Thomas Dean Rosinski

Michael Katz

Gerhard Sigl

Shailabh Nagar

Tathagata Chakraborti

Smruti Ranjan Sarangi

Raghavendra Uppinakuduru Udupa

Mangala Gowri, Iii

Techniques are provided for reinforcement learning software agents enhanced by external data. A reinforcement learning model supporting the software agent may be trained based on information obtained from one or more knowledge stores, such as online forums. The trained reinforcement learning model may be tested in an environment with limited connectivity to an external environment to meet performance criteria. The reinforcement learning software agent may be deployed with the tested and trained reinforcement learning model within an environment to autonomously perform actions to process requests.

Data augmented training of reinforcement learning software agent

Techniques that facilitate generating and executing an optimal dialogue strategy are provided. In one example, a system includes an information gain component and a question selector component. The information gain component estimates information gain data associated with first dialogue data of an information system. The information gain data is indicative of an amount of change in entropy associated with the first dialogue data. The question selector component selects between second dialogue data associated with a first question strategy and third dialogue data associated with a second question strategy based on the information gain data.

Generating and executing an optimal dialogue strategy

Embodiments relate to a system, program product, and method for automatically determining which activation data points in a neural model have been poisoned to erroneously indicate association with a particular label or labels. A neural network is trained using potentially poisoned training data. Each of the training data points is classified using the network to retain the activations of the last hidden layer, and segment those activations by the label of corresponding training data. Clustering is applied to the retained activations of each segment, and a cluster assessment is conducted for each cluster associated with each label to distinguish clusters with potentially poisoned activations from clusters populated with legitimate activations. The assessment includes executing a set of analyses and integrating the results of the analyses into a determination as to whether a training data set is poisonous based on determining if resultant activation clusters are poisoned.

Automatically determining poisonous attacks on neural networks

Techniques facilitating estimating and visualizing entity to agent collaboration to facilitate automated plan generation are provided. In one example, a computer-implemented method comprises generating, by a device operatively coupled to a processor, a plan based on receiving first input data associated with an instance model. The computer-implemented method also comprises generating, by the device, a revised plan based on receiving second input data, associated with a revised instance model, from an entity. Furthermore, the computer-implemented method comprises, tracking, by the device, a contribution of the entity as a function of a modification from the instance model to the revised instance model.

Estimating and visualizing collaboration to facilitate automated plan generation

Embodiments relate to a system, program product, and method for automatically determining which activation data points in a neural model have been poisoned to erroneously indicate association with a particular label or labels. A neural network is trained network using potentially poisoned training data. Each of the training data points is classified using the network to retain the activations of the last hidden layer, and segment those activations by the label of corresponding training data. Clustering is applied to the retained activations of each segment, and a cluster assessment is conducted for each cluster associated with each label to distinguish clusters with potentially poisoned activations from clusters populated with legitimate activations. The assessment includes analyzing, for each cluster, a distance of a median of the activations therein to medians of the activations in the labels.

Automatically determining whether an activation cluster contains poisonous data

A dialogue complexity assessment method, system, and computer program product including calculating a complexity utilizing domain-dependent terms and domain-independent terms of a dialogue, where the dialogue includes dialogue data from contact centers of service providers.

Assessing complexity of dialogs to streamline handling of service requests

Various embodiments are provided for integrating multiple domain learning and personalization in a dialog system for a user in a computing environment by a processor. One or more problem instances may be defined for multiple domains according to a problem instance template, identified user intent, links to one or more problem solvers associated with the multiple domains, or a combination thereof. A dialog plan may be determined to further define the one or more problem instances in response to user input. A solution may be provided to the user for the one or more problem instances.

Integrating multiple domain problem solving in a dialog system for a user

Various embodiments are provided for using a dialog system for integrating multiple domain learning and problem solving for a user in a computing environment by a processor. One or more problem instances may be defined for one or more selected domains in a multi-domain database according to a problem instance template, identified user intent, links to one or more problem solvers associated with the one or more selected domains, or a combination thereof. A dialog plan may be determined for the one or more problem instances using a dialog system associated with the multi-domain database, wherein each record in the multi-domain database corresponds to a selected database for the one or more selected domains. A solution may be provided to the user for the one or more problem instances. One or more preferences of a user may be learned according to the solution.

Using a dialog system for integrating multiple domain learning and problem solving

Techniques are provided for assigning a bias rating to a service based on bias and/or anti-bias testing. For example, unbiased data can be input to the service to, e.g., determine whether the service introduces elements of bias. As another example, biased data can be input to the service to determine whether the service provides elements of anti-bias. In one example, a computer-implemented method comprises selecting, by a device operatively coupled to a processor, source data configured according to a bias specification indicating a criterion that defines bias. The computer-implemented method can comprise determining whether bias exists in output data of the service and determining a bias rating based on the determining whether bias exists in the output data. The computer-implemented method can further comprise assigning the bias rating to the service.

Assigning bias ratings to services

Embodiments for learning personalized actionable domain models by a processor. A domain model may be generated according to a plurality of actions, extracted from one or more online data sources, of a plurality of cluster representatives. The plurality of actions achieve a goal. A hierarchical action model may be generated based on probabilities of the domain model and the plurality of actions. The hierarchical action model comprises a sequence of actions of the plurality of actions for achieving the goal. The hierarchical action model may be personalized by filtering to a selected set of actions according to weighted actions of the plurality of actions.

Learning personalized actionable domain models

Embodiments for building domain models from dialog interactions by a processor. A domain knowledge may be elicited from one or more dialog interactions with one or more users according to one or more dialog strategies. One or more domain models may be built and/or enhanced according to the domain knowledge.

Building domain models from dialog interactions

Methods and arrangements for assessing impact of an event on a network. An event notification is received relative to traffic in the network. Metadata are extracted from the notification, and the extracted metadata are converted into a predetermined format for processing. The converted metadata are received, and an impact of the event on traffic in the network is determined. Other variants and embodiments are broadly contemplated herein.

Extracting events and assessing their impact on a transportation network

A computer-implemented dynamic road stretch dividing method, the method comprising: determining a current lane distribution of partitions of a road stretch; calculating a new lane distribution of the road stretch to ameliorate traffic based on a pragmatic factor; changing an alignment of the partitions of the current lane distribution to obtain the new lane distribution; and updating the pragmatic factor based on at least one of an external policy and a constraint input by a user.

Dynamic road width division for adaptive road space utilization

Methods and arrangements for displaying results of one or more query searches. Query search results are obtained and there are determined, from the query search results, a best object type, a best instance of the best object type and one or more alternative object types, wherein determining a best object type includes consulting a collection of domain ontologies. The search results are displayed on a single screen via showing the best object type, the best instance of the best object type and the one or more alternative object types. Other variants and embodiments are broadly contemplated herein.

Compact visualization of query search results

Growing community of inventors

Noida, India

Biplav Srivastava