Fig 1 - uploaded by Maryam S Mirian
Content may be subject to copyright.
Source publication
![Fig. 4. . Examples of images in each of 9 segments of ACB (from [12]...](profile/Maryam-Mirian/publication/224120322/figure/fig4/AS:302870202208259@1449221291151/Examples-of-images-in-each-of-9-segments-of-ACB-from-12-downloaded-from-16_Q320.jpg)
We present a general mathematical description of the top-down attention control problem. Three important components are identified in the model: context extraction, attention focus and decision making. The context gives a coarse blurry representation of the whole input; the attention module models the focus of attention on a limited part of input,...
Context in source publication
Context 1
... modeling of and confirms the “coarse to fine” hypothesis [4], e.g. the sufficient information needed to decide whether a scene belongs to a natural or manmade category, can be obtained by a coarse view, and a finer categorization requires more detailed processing of specific features. After the agent has performed the motory action, for the sake of simplicity we assume that its mental state will become i.e. . Moreover, we assume that the agent’s real state in the environment does not change during the attention shifts between subsequent motory action selections. A schematic view of the above description of attention control scenario is given in Fig. 1. To this point, , we have introduced functions context extraction ݂ , attention shift ݂ , decision making ݂ , and mental state update ݂ ெௌ . Perceiving information from the environment is modeled by ݂ and ݂ . Therefore, the constraints that the agent’s processing limits limit impose on receiving information from the environment are modeled by these two functions. Context ontext extraction from sensory input requires simpler computations than processing the local focused attended input. As a result, result the load of operations done on each dimension of the sensory input is lower in the context phase. Let’s denote this load by ݈ܽ݀ ௧௫௧ and ݈ܽ݀ ௧௧௧ in context extraction and attention shift phases respectively. Thus, we would have: ݈ܽ݀ ௧௫௧ ݈ܽ݀ ௧௧௧ (6) and ...
Similar publications
This paper addresses the pollution detection problem by using a camera and analyzing the pictures. A camera is used to record visual scenes around complex plants. Then several signals are computed to describe the pictures. Our aim is to detect among the various clouds if there are polluting smokes. We assume in this paper that the signals are usefu...
In this paper, a novel relevance feedback is proposed to capture the user's query intention in large remote sensing databases. In order to increase the retrieval precision and reduce the retrieval time, the proposed relevance feedback approach firstly pre-selects a set of candidate images using semantic score functions as well as code co-occurrence...
Citations
... In other words, attention control can be described as a separate skill to concentrate [1]. This method acts like an active evaluation structure, which filters the part of input information that have, more important and related data to the action in progress [4]. The type of evaluation and filtering on the data is relevant to the robot's current situation and environment that a robot is surveying on, in that moment [4]. ...
... This method acts like an active evaluation structure, which filters the part of input information that have, more important and related data to the action in progress [4]. The type of evaluation and filtering on the data is relevant to the robot's current situation and environment that a robot is surveying on, in that moment [4]. ...
... One of the requirements in robotic applications is controlling attention on some special inputs to limit other processes and focus on desired task and inputs. This is more important when system actuators, inputs and sensors grow up in number [4], [5]. Without attention algorithms, it is impossible to process all observable information that is received per second [5]. ...
... In each robot learning situation, we are facing a learning problem that includes two separate parts [8]: ...
... HOG or CENTRIS on image). This sensor modeling named virtual sensor [8], so we formulate the information space as Eq.2. ...
Attention control is one of the best ways to reduce information resources and processing. Discontinuous modeling has been used in attention control and has proved some advantages of attention control. In this paper we present attention control architecture based on continuous modeling for mobile robot platforms. By using fuzzy neural network we construct efficient attention control which is capable of decreasing sensors sampling rate and also choosing the most efficient set of sensors. We also build a novel method for gathering information to construct fuzzy neural networks. We experimentally proved that fuzzy neural networks are very convenient ways for attention control. By using this method which changes sampling rate of robot sensors, consumption of energy reduces slightly. This novel framework is implemented on Festo Robotino® mobile robot platform and results show efficiency of this attention control method which can select the best sensors during each task.
... On the other hand, inter-sensor partitioning seems to be a more demanding task. Inter-sensor partitioning can be performed automatically by an optimization method [22] or based on methods such as agglomerative clustering [30]. The other possible solution is to hand-design the partitioning. ...
Rapid increase in the size and complexity of sensory systems demands for attention control in real world robotic tasks. However, attention control and the task are often highly interlaced which demands for interactive learning. In this paper, a framework called METAL mixture-of-experts task and attention learning is proposed to cope with this complex learning problem. METAL consists of three consecutive learning phases, where the first two phases provide an initial knowledge about the task, while in the third phase the attention control is learned concurrently with the task. The mind of the robot is composed of a set of tiny agents learning and acting in parallel in addition to an attention control learning ACL agent. Each tiny agent provides the ACL agent with some partial knowledge about the task in the form of its decision preference-called policy as well. The ACL agent in the third phase learns how to make the final decision by attending the least possible number of tiny agents. It acts on a continuous decision space which gives METAL the ability to integrate different sources of knowledge with ease. A Bayesian continuous RL method is utilized at both levels of learning on perceptual and decision spaces. Implementation of METAL on an E-puck robot in a miniature highway driving task along with farther simulation studies in Webots™ environment verify the applicability and effectiveness of the proposed framework, where a smooth driving behavior is shaped. It is also shown that even though the robot has learned to discard some sensory data, probability of raising aliasing in the decision space is very low, which means that the robot can learn the task as well as attention control simultaneously.
