Budget-Aware Adapters for Multi-Domain Learning

被引:23
作者
Berriel, Rodrigo [1 ,5 ]
Lathuiliere, Stephane [2 ]
Nabi, Moin [3 ]
Klein, Tassilo [3 ]
Oliveira-Santos, Thiago [1 ]
Sebe, Nicu [2 ]
Ricci, Elisa [2 ,4 ]
机构
[1] Univ Fed Espirito Santo, LCAD, Vitoria, ES, Brazil
[2] Univ Trento, DISI, Trento, Italy
[3] SAP ML Res, Trento, Italy
[4] Fdn Bruno Kessler, Povo, Italy
[5] Univ Trento, DISI, MHUG, Trento, Italy
来源
2019 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2019) | 2019年
关键词
D O I
10.1109/ICCV.2019.00047
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Multi-Domain Learning (MDL) refers to the problem of learning a set of models derived from a common deep architecture, each one specialized to perform a task in a certain domain (e.g., photos, sketches, paintings). This paper tackles MDL with a particular interest in obtaining domain-specific models with an adjustable budget in terms of the number of network parameters and computational complexity. Our intuition is that, as in real applications the number of domains and tasks can be very large, an effective MDL approach should not only focus on accuracy but also on having as few parameters as possible. To implement this idea we derive specialized deep models for each domain by adapting a pre-trained architecture but, differently from other methods, we propose a novel strategy to automatically adjust the computational complexity of the network. To this aim, we introduce Budget-Aware Adapters that select the most relevant feature channels to better handle data from a novel domain. Some constraints on the number of active switches are imposed in order to obtain a network respecting the desired complexity budget. Experimentally, we show that our approach leads to recognition accuracy competitive with state-of-the-art approaches but with much lighter networks both in terms of storage and computation.
引用
收藏
页码:382 / 391
页数:10
相关论文
共 39 条
[1]   Memory Aware Synapses: Learning What (not) to Forget [J].
Aljundi, Rahaf ;
Babiloni, Francesca ;
Elhoseiny, Mohamed ;
Rohrbach, Marcus ;
Tuytelaars, Tinne .
COMPUTER VISION - ECCV 2018, PT III, 2018, 11207 :144-161
[2]  
[Anonymous], 2018, CVPR
[3]  
[Anonymous], 2016, INT J DIGITAL ART HI
[4]  
[Anonymous], 2015, Arxiv.Org, DOI DOI 10.3389/FPSYG.2013.00124
[5]  
[Anonymous], 2014, P IEEE CVPR
[6]  
[Anonymous], 2016, NEURAL INFORM PROCES
[7]  
[Anonymous], 2018, IEEE T PATTERN ANAL
[8]   Towards Open Set Deep Networks [J].
Bendale, Abhijit ;
Boult, Terrance E. .
2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2016, :1563-1572
[9]  
BILEN H, 2016, PROC CVPR IEEE, P3034, DOI DOI 10.1109/CVPR.2016.331
[10]  
Bissacco A., 2011, NEUR INF PROC SYST N