Differences
This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revision | Next revisionBoth sides next revision |
| pub:teaching:courses:rnd [2018/05/30 10:08] – [2018-04-25] sbk | pub:teaching:courses:rnd [2018/05/30 10:09] – [2018-04-25] sbk |
|---|
| |
| **Contents:** | **Contents:** |
| - Team Introduction | - **Team Introduction** |
| - Experiment #1 - 'Deep Learning based Natural Language Search' \\ In this talk, we will provide details of a Deep Learning based solution (DrQA) for Natural Language Search over factory machine manuals. The proposed architecture allows identifying the precise answer span corresponding to a query, which improves over the paragraph level answers enabled by pure statistical approaches. | - **Experiment #1 - 'Deep Learning based Natural Language Search'** \\ In this talk, we will provide details of a Deep Learning based solution (DrQA) for Natural Language Search over factory machine manuals. The proposed architecture allows identifying the precise answer span corresponding to a query, which improves over the paragraph level answers enabled by pure statistical approaches. |
| - Experiment #2 - 'Planogram - Shelf Compliance at Point of Sale' \\ In this talk we will provide details of a Deep Learning based solution for shelf compliance image recognition use-case at PMI Points-of-Sales (PoS). The experimental architecture consists of two stages, first one detects all positions on the shelf (empty positions, our and non PMI products). The second one is executing classification of detected positions and provides a type of each. System finds all PMI products and their types, non-PMI products, empty slots, counts them and verifies if everything is placed correctly due to PMI requirements. | - **Experiment #2 - 'Planogram - Shelf Compliance at Point of Sale'** \\ In this talk we will provide details of a Deep Learning based solution for shelf compliance image recognition use-case at PMI Points-of-Sales (PoS). The experimental architecture consists of two stages, first one detects all positions on the shelf (empty positions, our and non PMI products). The second one is executing classification of detected positions and provides a type of each. System finds all PMI products and their types, non-PMI products, empty slots, counts them and verifies if everything is placed correctly due to PMI requirements. |
| - Experiment #3 - 'Helpdesk Ticket Classification' \\ In this talk we will provide details about neural network based solution for helpdesk ticket classification. The experimental network architecture consists of a mix of fully connected layers on top of number of keyword occurrences and a stack of one dimensional convolutions over word vector embeddings for the whole text. | - **Experiment #3 - 'Helpdesk Ticket Classification'** \\ In this talk we will provide details about neural network based solution for helpdesk ticket classification. The experimental network architecture consists of a mix of fully connected layers on top of number of keyword occurrences and a stack of one dimensional convolutions over word vector embeddings for the whole text. |
| - Discussion & Closing | - **Discussion & Closing** |
| |
| **Speakers’ Biograms** | **Speakers’ Biograms** |
