Long-Term Ensemble Learning for Cross-Season Visual Place Classification

Author(s):  
Xiaoxiao Fei ◽  
◽  
Kanji Tanaka ◽  
Yichu Fang ◽  
Akitaka Takayama

This paper addresses the problem of cross-season visual place classification (VPC) from the novel perspective of long-term map learning. Our goal is to enable transfer learning efficiently from one season to the next, at a small constant cost, and without wasting the robot’s available long-term-memory by memorizing very large amounts of training data. To achieve a good tradeoff between generalization and specialization abilities, we employ an ensemble of deep convolutional neural network (DCN) classifiers and consider the task of scheduling (when and which classifiers to retrain), given a previous season’s DCN classifiers as the sole prior knowledge. We present a unified framework for retraining scheduling and we discuss practical implementation strategies. Furthermore, we address the task of partitioning a robot’s workspace into places to define place classes in an unsupervised manner, as opposed to using uniform partitioning, so as to maximize VPC performance. Experiments using the publicly available NCLT dataset revealed that retraining scheduling of a DCN classifier ensemble is crucial in achieving a good balance between generalization and specialization. Additionally, it was found that the performance is significantly improved when using planned scheduling.

2016 ◽  
Vol 28 (1) ◽  
pp. 187-197 ◽  
Author(s):  
J. Benjamin Hutchinson ◽  
Sarah S. Pak ◽  
Nicholas B. Turk-Browne

A key task for the brain is to determine which pieces of information are worth storing in memory. To build a more complete representation of the environment, memory systems may prioritize new information that has not already been stored. Here, we propose a mechanism that supports this preferential encoding of new information, whereby prior experience attenuates neural activity for old information that is competing for processing. We evaluated this hypothesis with fMRI by presenting a series of novel stimuli concurrently with repeated stimuli at different spatial locations in Experiment 1 and from different visual categories (i.e., faces and scenes) in Experiment 2. Subsequent memory for the novel stimuli could be predicted from the reduction in activity in ventral temporal cortex for the accompanying repeated stimuli. This relationship was eliminated in control conditions where the competition during encoding came from another novel stimulus. These findings reveal how prior experience adaptively guides learning toward new aspects of the environment.


2017 ◽  
Vol 62 (1) ◽  
Author(s):  
Mona Helmchen

AbstractIn the present article and in a theoretical frame based in cognitive linguistics, we will justify and analyse some of the translational processes that have taken place during the translation of the novel “La Tesis de Nancy” (1962). The human being constantly perceives a multitude of impressions which operate as mental impulses and which activate parts of the long-term memory to facilitate the understanding of new information. Apart from the development of mental spaces, the metaphorical thinking plays an important role in the creation of meaning and in the understanding of reality. A professional translator has to consider that an awareness of such processes can be an inspirational source for a creative and functional translation.


2013 ◽  
Vol 105 ◽  
pp. 174-185 ◽  
Author(s):  
Shannon J. Moore ◽  
Kaivalya Deshpande ◽  
Gwen S. Stinnett ◽  
Audrey F. Seasholtz ◽  
Geoffrey G. Murphy

eLife ◽  
2017 ◽  
Vol 6 ◽  
Author(s):  
Alan Jung Park ◽  
Robbert Havekes ◽  
Xiuping Fu ◽  
Rolf Hansen ◽  
Jennifer C Tudor ◽  
...  

Long-lasting forms of synaptic plasticity and memory require de novo protein synthesis. Yet, how learning triggers this process to form memory is unclear. Translin/trax is a candidate to drive this learning-induced memory mechanism by suppressing microRNA-mediated translational silencing at activated synapses. We find that mice lacking translin/trax display defects in synaptic tagging, which requires protein synthesis at activated synapses, and long-term memory. Hippocampal samples harvested from these mice following learning show increases in several disease-related microRNAs targeting the activin A receptor type 1C (ACVR1C), a component of the transforming growth factor-β receptor superfamily. Furthermore, the absence of translin/trax abolishes synaptic upregulation of ACVR1C protein after learning. Finally, synaptic tagging and long-term memory deficits in mice lacking translin/trax are mimicked by ACVR1C inhibition. Thus, we define a new memory mechanism by which learning reverses microRNA-mediated silencing of the novel plasticity protein ACVR1C via translin/trax.


2016 ◽  
Vol 39 ◽  
Author(s):  
Mary C. Potter

AbstractRapid serial visual presentation (RSVP) of words or pictured scenes provides evidence for a large-capacity conceptual short-term memory (CSTM) that momentarily provides rich associated material from long-term memory, permitting rapid chunking (Potter 1993; 2009; 2012). In perception of scenes as well as language comprehension, we make use of knowledge that briefly exceeds the supposed limits of working memory.


2020 ◽  
Vol 29 (4) ◽  
pp. 710-727
Author(s):  
Beula M. Magimairaj ◽  
Naveen K. Nagaraj ◽  
Alexander V. Sergeev ◽  
Natalie J. Benafield

Objectives School-age children with and without parent-reported listening difficulties (LiD) were compared on auditory processing, language, memory, and attention abilities. The objective was to extend what is known so far in the literature about children with LiD by using multiple measures and selective novel measures across the above areas. Design Twenty-six children who were reported by their parents as having LiD and 26 age-matched typically developing children completed clinical tests of auditory processing and multiple measures of language, attention, and memory. All children had normal-range pure-tone hearing thresholds bilaterally. Group differences were examined. Results In addition to significantly poorer speech-perception-in-noise scores, children with LiD had reduced speed and accuracy of word retrieval from long-term memory, poorer short-term memory, sentence recall, and inferencing ability. Statistically significant group differences were of moderate effect size; however, standard test scores of children with LiD were not clinically poor. No statistically significant group differences were observed in attention, working memory capacity, vocabulary, and nonverbal IQ. Conclusions Mild signal-to-noise ratio loss, as reflected by the group mean of children with LiD, supported the children's functional listening problems. In addition, children's relative weakness in select areas of language performance, short-term memory, and long-term memory lexical retrieval speed and accuracy added to previous research on evidence-based areas that need to be evaluated in children with LiD who almost always have heterogenous profiles. Importantly, the functional difficulties faced by children with LiD in relation to their test results indicated, to some extent, that commonly used assessments may not be adequately capturing the children's listening challenges. Supplemental Material https://doi.org/10.23641/asha.12808607


2011 ◽  
Vol 70 (1) ◽  
pp. 35-39 ◽  
Author(s):  
Muriel Fanget ◽  
Catherine Thevenot ◽  
Caroline Castel ◽  
Michel Fayol

In this study, we used a paradigm recently developed ( Thevenot, Fanget, & Fayol, 2007 ) to determine whether 10-year-old children solve simple addition problems by retrieval of the answer from long-term memory or by calculation procedures. Our paradigm is unique in that it does not rely on reaction times or verbal reports, which are known to potentially bias the results, especially in children. Rather, it takes advantage of the fact that calculation procedures degrade the memory traces of the operands, so that it is more difficult to recognize them when they have been involved in the solution of an addition problem by calculation rather than by retrieval. The present study sharpens the current conclusions in the literature and shows that, when the sum of addition problems is up to 10, children mainly use retrieval, but when it is greater than 10, they mainly use calculation procedures.


2010 ◽  
Vol 24 (4) ◽  
pp. 249-252 ◽  
Author(s):  
Márk Molnár ◽  
Roland Boha ◽  
Balázs Czigler ◽  
Zsófia Anna Gaál

This review surveys relevant and recent data of the pertinent literature regarding the acute effect of alcohol on various kinds of memory processes with special emphasis on working memory. The characteristics of different types of long-term memory (LTM) and short-term memory (STM) processes are summarized with an attempt to relate these to various structures in the brain. LTM is typically impaired by chronic alcohol intake but according to some data a single dose of ethanol may have long lasting effects if administered at a critically important age. The most commonly seen deleterious acute effect of alcohol to STM appears following large doses of ethanol in conditions of “binge drinking” causing the “blackout” phenomenon. However, with the application of various techniques and well-structured behavioral paradigms it is possible to detect, albeit occasionally, subtle changes of cognitive processes even as a result of a low dose of alcohol. These data may be important for the consideration of legal consequences of low-dose ethanol intake in conditions such as driving, etc.


Author(s):  
Angela A. Manginelli ◽  
Franziska Geringswald ◽  
Stefan Pollmann

When distractor configurations are repeated over time, visual search becomes more efficient, even if participants are unaware of the repetition. This contextual cueing is a form of incidental, implicit learning. One might therefore expect that contextual cueing does not (or only minimally) rely on working memory resources. This, however, is debated in the literature. We investigated contextual cueing under either a visuospatial or a nonspatial (color) visual working memory load. We found that contextual cueing was disrupted by the concurrent visuospatial, but not by the color working memory load. A control experiment ruled out that unspecific attentional factors of the dual-task situation disrupted contextual cueing. Visuospatial working memory may be needed to match current display items with long-term memory traces of previously learned displays.


Sign in / Sign up

Export Citation Format

Share Document