BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//jEvents 2.0 for Joomla//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
UID:86svfdcq7djbnn7884ndscbhi8@google.com
CATEGORIES:RTportal.ru
CREATED:20170606T092330
SUMMARY::christmas_tree: Christmas Colloquium on Computer Vision
LOCATION:Scolkovo, Moscow Oblast, Russia
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:Six researchers will present their recent works from ECCV’14, CVPR’15, ICCV
 ’15, ICML’15, ICLR’16(submitted) of this year. The language of the talk wil
 l be chosen by the speakers.\n\nCenter for Data-Intensive Science and Engin
 eering (CDISE) of Skoltech invites you to participate (pre-registration nee
 ded)!\nFor the pre-registration, please send email with the subject ‘CCCV r
 egistration’ and your name to Адрес электронной почты защищен от спам-ботов
 . Для просмотра адреса в вашем браузере должен быть включен Javascript.
			
 	document.getElementById('cloakcb721ffa56a11ee14d26ee37da50b2c4').innerHTML
  = '';
				var prefix = 'ma' + 'il' + 'to';
				var path = '
 hr' + 'ef' + '=';
				var addycb721ffa56a11ee14d26ee37da50b2c4 = 'ev&#
 101;nts' + '@';
				addycb721ffa56a11ee14d26ee37da50b2c4 = addycb721ffa
 56a11ee14d26ee37da50b2c4 + 'skoltech' + '.' + 'ru';
				
 var addy_textcb721ffa56a11ee14d26ee37da50b2c4 = 'events' + '@
 ' + 'skoltech' + '.' + 'ru';document.getElementById('clo
 akcb721ffa56a11ee14d26ee37da50b2c4').innerHTML += ''+addy_textcb721ffa56a11
 ee14d26ee37da50b2c4+'';
		 (CC Адрес электронной почты защищен от спам-бото
 в. Для просмотра адреса в вашем браузере должен быть включен Javascript.
		
 		document.getElementById('cloak95abcd59b69cf6709224755674eeaca2').innerHTM
 L = '';
				var prefix = 'ma' + 'il' + 'to';
				var path = 
 'hr' + 'ef' + '=';
				var addy95abcd59b69cf6709224755674eeaca2 = 'lem
 pitsky' + '@';
				addy95abcd59b69cf6709224755674eeaca2 = addy95ab
 cd59b69cf6709224755674eeaca2 + 'skoltech' + '.' + 'ru';

 				var addy_text95abcd59b69cf6709224755674eeaca2 = 'lempitsky' +
  '@' + 'skoltech' + '.' + 'ru';document.getElementBy
 Id('cloak95abcd59b69cf6709224755674eeaca2').innerHTML += ''+addy_text95abcd
 59b69cf6709224755674eeaca2+'';
		)\n\nSCHEDULE:\n\n14:00 — 14:30\nAsya Pent
 ina, IST Austria, Vienna\nTitle: Curriculum learning of multiple tasks\nAbs
 tract: Sharing information between multiple tasks enables algorithms to ach
 ieve good generalization performance even from small amounts of training da
 ta. However, in a realistic scenario of multi-task learning not all tasks a
 re equally related to each other, hence it could be advantageous to transfe
 r information only between the most related tasks.\nIn this work we propose
  an approach that processes multiple tasks in a sequence with sharing betwe
 en subsequent tasks instead of solving all tasks jointly. Subsequently, we 
 address the question of curriculum learning of tasks, i.e. finding the best
  order of tasks to be learned. Our approach is based on a generalisation bo
 und criterion for choosing the task order that optimises the average expect
 ed classification performance over all tasks.Our experimental results show 
 that learning multiple related tasks sequentially can be more effective tha
 n learning them jointly, the order in which tasks are being solved affects 
 the overall performance, and that our model is able to automatically discov
 er a favourable order of tasks.\n\n14:30 — 15:00\nSergey Zagoruyko, École d
 e Ponts ParisTech (talk based on the work at Facebook AI Research)\nTitle: 
 FAIRCNN MSCOCO object detection/segmentation challenge submission.\nAbstrac
 t: Our approach is built on DeepMask proposals fed into the Fast R-CNN pipe
 line. The DeepMask proposals have been substantially improved to encourage 
 proposal diversity and mask quality. We also augmented our CNN classifier w
 ith a novel foveal structure, skip-connections, an improved cost function t
 hat encourages better localization, and a few additional modifications. Fin
 ally we utilize ensembling (model and inference) to further improve perform
 ance.\n\n15:00 — 15:15\nBreak\n\n15:15 — 15:45\nAnton Osokin, INRIA/École N
 ormale Supérieure, Paris\nTitle: Context-aware CNNs for person head detecti
 on\nAbstract: Person detection is a key problem for many computer vision ta
 sks. While face detection has reached maturity, detecting people under a fu
 ll variation of camera view-points, human poses, lighting conditions and oc
 clusions is still a difficult challenge. In this work we focus on detecting
  human heads in natural scenes. Starting from the recent local R-CNN object
  detector, we extend it with two types of contextual cues. First, we levera
 ge person-scene relations and propose a Global CNN model trained to predict
  positions and scales of heads directly from the full image. Second, we exp
 licitly model pairwise relations among objects and train a Pairwise CNN mod
 el using a structured-output surrogate loss. The Local, Global and Pairwise
  models are combined into a joint CNN framework. To train and test our full
  model, we introduce a large dataset composed of 369,846 human heads annota
 ted in 224,740 movie frames. We evaluate our method and demonstrate improve
 ments of person head detection against several recent baselines in three da
 tasets. We also show improvements of the detection speed provided by our mo
 del.\n\n15:45 — 16:15\nDanila Potapov, INRIA-LEAR, Grenoble\nTitle: Categor
 y-specific video summarization\nAbstract: In large video collections with c
 lusters of typical categories, such as “birthday party” or “flash-mob”, cat
 egory-specific video summarization can produce higher quality video summari
 es than unsupervised approaches that are blind to the video category.\nGive
 n a video from a known category, our approach first efficiently performs a 
 temporal segmentation into semantically-consistent segments, delimited not 
 only by shot boundaries but also general change points. Then, equipped with
  an SVM classifier, our approach assigns importance scores to each segment.
  The resulting video assembles the sequence of segments with the highest sc
 ores. The obtained video summary is therefore both short and highly informa
 tive. Experimental results on videos from the multimedia event detection (M
 ED) dataset of TRECVID’11 show that our approach produces video summaries w
 ith higher relevance than the state of the art.\n\n16:15 — 16:30\nBreak\n\n
 16:30 — 17:00\nMichael Figurnov, Skoltech, Moscow\nTitle: PerforatedCNNs: A
 cceleration through Elimination of Redundant Convolutions\nAbstract: We pro
 pose a novel approach to reduce the computational cost of evaluation of con
 volutional neural networks, a factor that has hindered their deployment in 
 low-power devices such as mobile phones. Inspired by the loop perforation t
 echnique from source code optimization, we speed up the bottleneck convolut
 ional layers by skipping their evaluation in some of the spatial positions.
  We propose and analyze several strategies of choosing these positions. Our
  method allows to reduce the evaluation time of modern convolutional neural
  networks by 50% with a small decrease in accuracy. More details can be fou
 nd in our ICLR 2016 submission: http://arxiv.org/pdf/1504.08362v2.pdf\n\n17
 :00 — 17:30\nYaroslav Ganin, Skoltech, Moscow\nTitle: Unsupervised Domain A
 daptation by Backpropagation\nAbstract: Top-performing deep architectures a
 re trained on massive amounts of labeled data. In the absence of labeled da
 ta for a certain task, domain adaptation often provides an attractive optio
 n given that labeled data of similar nature but from a different domain (e.
 g. synthetic images) are available. Here, we propose a new approach to doma
 in adaptation in deep architectures that can be trained on large amount of 
 labeled data from the source domain and large amount of unlabeled data from
  the target domain (no labeled target-domain data is necessary). As the tra
 ining progresses, the approach promotes the emergence of “deep” features th
 at are (i) discriminative for the main learning task on the source domain a
 nd (ii) invariant with respect to the shift between the domains. We show th
 at this adaptation behaviour can be achieved in almost any feed-forward mod
 el by augmenting it with few standard layers and a simple new gradient reve
 rsal layer. The resulting augmented architecture can be trained using stand
 ard backpropagation. Overall, the approach can be implemented with little e
 ffort using any of the deep-learning packages. The method performs very wel
 l in a series of image classification experiments, achieving adaptation eff
 ect in the presence of big domain shifts and outperforming previous state-o
 f-the-art on Office datasets. We also validate the approach for descriptor 
 learning task in the context of person re-identification application.\n\nht
 tp://sites.skoltech.ru/compvision/events/cccv2015/ This event was imported 
 from: https://rtportal.ru/index.php/kalendar-sobytij/eventdetail/680/-/chri
 stmas-tree-christmas-colloquium-on-computer-vision?tmpl=component
DTSTAMP:20260812T230915Z
DTSTART;TZID=Europe/Moscow:20151228T140000
DTEND;TZID=Europe/Moscow:20151228T180000
SEQUENCE:0
TRANSP:OPAQUE
END:VEVENT
END:VCALENDAR