Abstract

Abstract
Teaching industrial robots by demonstration can significantly decrease the repurposing costs of assembly lines worldwide. To achieve this goal, the robot needs to detect and track each component with high accuracy. To speedup the initial object recognition phase, the learning system can gather information from assembly manuals in order to identify which parts and tools are required for assembling a new product (avoiding exhaustive search in a large model database) and if possible also extract the assembly order and spatial relation between them. This paper presents a detailed analysis of the fine tuning of the Stanford Named Entity Recognizer for this text tagging task. Starting from the recommended configuration, it was performed 91 tests targeting the main features / parameters. Each test only changed a single parameter in relation to the recommend configuration, and its goal was to see the impact of the new configuration in the precision, recall and F1 metrics. This analysis allowed to fine tune the Stanford NER system, achieving a precision of 89.91%, recall of 83.51% and F1 of 84.69%. These results were retrieved with our new manually annotated dataset containing text with assembly operations for alternators, gearboxes and engines, which were written in a language discourse that ranges from professional to informal. The dataset can also be used to evaluate other information extraction and computer vision systems, since most assembly operations have pictures and diagrams showing the necessary product parts, their assembly order and relative spatial disposition. © 2017 IEEE.

Abstract
Search engines are evolving towards richer and stronger semantic approaches, focusing on entity-oriented tasks where knowledge bases have become fundamental. In order to support semantic search, search engines are increasingly reliant on robust information extraction systems. In fact, most modern search engines are already highly dependent on a well-curated knowledge base. Nevertheless, they still lack the ability to e ectively and automatically take advantage of multiple heterogeneous data sources. Central tasks include harnessing the information locked within textual content by linking mentioned entities to a knowledge base, or the integration of multiple knowledge bases to answer natural language questions. Combining text and knowledge bases is frequently used to improve search results, but it can also be used for the query-independent ranking of entities like events. In this work, we present a complete information extraction pipeline for the Portuguese language, covering all stages from data acquisition to knowledge base population. We also describe a practical application of the automatically extracted information, to support the ranking of upcoming events displayed in the landing page of an institutional search engine, where space is limited to only three relevant events. We manually annotate a dataset of news, covering event announcements from multiple faculties and organic units of the institution. We then use it to train and evaluate the named entity recognition module of the pipeline. We rank events by taking advantage of identified entities, as well as partOf relations, in order to compute an entity popularity score, as well as an entity click score based on implicit feedback from clicks from the institutional search engine. We then combine these two scores with the number of days to the event, obtaining a final ranking for the three most relevant upcoming events. © José Devezas and Sérgio Nunes

Abstract
In an information society, people expect to find answers to their questions quickly and with little effort. Sometimes, these answers are locked within textual documents, which often require a manual analysis, after being retrieved from the web using search engines. At FEUP InfoLab, we are researching graph-based models to index combined data (text and knowledge), with the goal of improving entity-oriented search effectiveness.

Abstract

Abstract
Many embedded applications process large amounts of data using regular computational kernels, amenable to acceleration by specialized hardware coprocessors. To reduce the significant design effort, the dedicated hardware may be automatically generated, usually starting from the application's source or binary code. This paper presents a moduloscheduled loop accelerator capable of executing multiple loops and a supporting toolchain. A generation/scheduling procedure, which fully relies on MicroBlaze instruction traces, produces accelerator instances, customized in terms of functional units and interconnections. The accelerators support integer and single-precision floating-point arithmetic, and exploit instruction-level parallelism, loop pipelining, and memory access parallelism via two read/write ports. A complete implementation of the proposed architecture is evaluated in a Virtex-7 device. Augmenting a MicroBlaze processor with a tailored accelerator achieves a geometric mean speedup, over software-only execution, of 6.61x for 13 floating-point kernels from the Livermore Loops set, and of 4.08x for 11 integer kernels from Texas Instruments' IMGLIB. The proposed customized accelerators are compared with ALU-based ones. The average specialized accelerator requires only 0.47x the number of field-programmable gate array slices of an accelerator with four ALUs. A geometric mean speedup of 1.78x over a four-issue very long instruction word (without floating-point support) was obtained for the integer kernels.