Paper of the day – Unsupervised Method for Video Action Segmentation Through Spatio-Temporal and Positional-Encoded Embeddings

 

#paperoftheday

Title: Unsupervised Method for Video Action Segmentation Through Spatio-Temporal and Positional-Encoded Embedding

Venue: ACM Multimedia Systems Conference (2022)

Authors: Guilherme de A. P. Marques, Antonio José G. Busson, Álan Lívio V. Guedes, Julio Cesar Duarte, Sérgio Colcher

Abstract: Action segmentation consists of temporally segmenting a video and labeling each segmented interval with a specific action label. In this work, we propose a novel action segmentation method that requires no prior video analysis and no annotated data. Our method involves extracting spatio-temporal features from videos using a pre-trained deep network. Data is then transformed using a positional encoder, and finally a clustering algorithm is applied, where each produced cluster presumably corresponds to a different single and distinguishable action. In experiments, we show that our method produces competitive results on the Breakfast and Inria Instructional Videos dataset benchmarks.

More in: https://doi.org/10.1145/3524273.3528187

Paper of the day – Clustering problems

#paperoftheday

Title:
On the price of explainability for some clustering problems Venue: International Conference on Machine Learning (ICML), 2021

Authors:
Eduardo Sany Laber, Lucas Murtinho

Abstract:
The price of explainability for a clustering task can be defined as the unavoidable loss, in terms of the objective function, if we force the final partition to be explainable. Here, we study this price for the following clustering problems: k-means, k-medians, k-centers and maximum-spacing. We provide upper and lower bounds for a natural model where explainability is achieved via decision trees. For the k-means and k-medians problems our upper bounds improve those obtained by [Dasgupta et. al, ICML 20] for low dimensions. Another contribution is a simple and efficient algorithm for building explainable clusterings for the k-means problem. We provide empirical evidence that its performance is better than the current state of the art for decision-tree based explainable clustering.

More in:
http://proceedings.mlr.press/v139/laber21a.html

Paper of the day – Computational narrative blending based on planning

#paperoftheday

Title:
Computational Narrative Blending Based on Planning

Venue:
International Conference on Entertainment Computing (ICEC), 2021

Authors:
Edirlei Soares de Lima, Bruno Feijó, Antonio L. Furtado

Abstract:
Inspired by conceptual blending models and considering plot generation as a plan- generation problem, this paper proposes a robust method that reuses existing stories to generate new narrative variants. This method generates variants that combine episodes extracted and adapted from different stories that share the same narrative structure. By combining a plan validation algorithm with a basic narrative structure, our method guarantees the logical coherence and general plot structure of the generated narratives. We also propose a new tool to assist amateur/professional writers to visualize all narrative variants created from a set of existing stories. Our experiments created novel, coherent and structured narratives by blending and adapting episodes from old chivalry romance pieces of work and some modern adventure videogames.

More in:
https://doi.org/10.1007/978-3-030-89394-1_22

Paper of the day – Analysis of a many-objective optimization

#paperoftheday

Tittle:

Analysis of a many-objective optimization approach for identifying microservices from legacy systems

Venue:

Empirical Software Engineering volume 27, Article number: 51 (2022)

Authors: Wesley K. G. Assunção, Thelma Elita Colanzi, Luiz Carvalho, Alessandro Garcia, Juliana Alves Pereira, Maria Julia de Lima, Carlos Lucena

Abstract:

The expensive maintenance of legacy systems leads companies to migrate such systems to modern architectures. Microservice architectural style has become a trend to modernize monolithic legacy systems. A microservice architecture consists of small, autonomous, and highly-independent services communicating by using lightweight network protocols. To support the designing of microservice architectures, recent studies have proposed either single or multi-objective approaches. In order to improve the effectiveness of existing approaches, we introduced toMicroservices that is a many-objective search-based approach to aid the identification of boundaries among services. In previous studies, we have focused on a qualitative evaluation of the applicability and adoption of the proposed approach from a practical point of view, thus the optimization process itself has not been investigated in depth. In this paper, we extend our previous work by performing a more in-depth analysis of our many-objective approach for microservice identification. We compare our approach against a baseline approach based on a random search using a set of performance indicators widely used in the literature of many-objective optimization. Our results are validated through a real-world case study. The study findings reveal that (i) the criteria optimized by our approach are interdependent and conflicting; and (ii) all candidate solutions lead to better performance indicators in comparison to random search. Overall, the proposed many-objective approach for microservice identification yields promising results, which shed light on insights for further improvements.

More in:

https://link.springer.com/article/10.1007/s10664-021-10049-7

Professores do DI estão entre os mais influentes do Brasil, aponta ranking

Campus da PUC-Rio

Professores Lucena, Garcia e Schwabe estão entre os 33 pesquisadores de maior destaque na área, segundo a Guide2Research

 

Três professores do Departamento de Informática (DI) da PUC-Rio estão entre os 33 pesquisadores mais influentes em Ciência da Computação e Eletrônica no Brasil, segundo o ranking da Guide2Research, principal portal de pesquisa na área. O professor Carlos José Pereira de Lucena aparece na 11ª posição, o professor Alessandro Garcia  na 19ª e o professor aposentado Daniel Schwabe está na 30ª colocação. 

 

A classificação leva em conta os cientistas que tenham índice H do Google Scholar a partir de 40 e verifica suas publicações no Digital Bibliography & Library Project (DBLP). Também são avaliadas premiações e bolsas de cada pesquisador. Esta edição do ranking se baseou em dados coletados até 10 de maio de 2021.

 

Professor Carlos José Pereira de Lucena

Um dos grandes nomes e fundador do DI, Carlos Lucena tem uma trajetória profissional singular. Segundo a tabela, o diretor do Laboratório de Engenharia de Software do DI tem 345 artigos registrados no DBLP, 12,250 citações e índice H = 52. Lucena é também ganhador do prêmio ACM Fellow (2013) e do ACM Distinguished Member (2009). Seu artigo de maior destaque no Google Scholar, “Modularizing design patterns with aspects: a quantitative study”(2006), conta com 434 citações. Em seguida vem o  “On the reuse and maintenance of aspect-oriented software: An assessment framework”, com 365 menções.

 

Ainda de acordo com o ranking, o professor associado do DI Alessandro Garcia, que também atua no campo de Engenharia de Software, tem 253 artigos  registrados no DBLP, 9,126 citações e índice H = 46. Vários de seus trabalhos receberam distinções nas principais conferências internacionais da sua área, como o “ACM Distinguished Paper Award” do ICSE 2014. Atualmente, Garcia trabalha em parceria com grupos de pesquisa internacionais nos EUA, Inglaterra, Alemanha e Argentina. 

 

Professor Alessandro Garcia

O professor aposentado Daniel Schwabe aparece na lista com 149 artigos registrados no DBLP, 8,685 citações e índice H = 40. No Google Scholar, seu artigo mais relevante é o “HDM—a model-based approach to hypertext application design”(1993), que chega a 1105 citações. Em seguida está o trabalho “An object oriented approach to web-based applications design”(1998), citado 666 vezes.