In order to elucidate the neural mechanisms involved in the perception of the three-dimensional (3D) orientation of a surface, we trained monkeys to discriminate the 3D orientation of a surface from binocular disparity cues using a Go/No-go type delayed-matching-to-sample (DMTS) task and examined the properties of the surface-orientation-selective (SOS) neurons. We recorded 57 SOS neurons from the caudal part of the lateral bank of the intraparietal sulcus (area CIP) of three hemispheres of two Japanese monkeys (Macaca fuscata). We tested 29 of 57 SOS neurons using the square plate of a solid figure stereogram (SFS) and random-dot stereogram (RDS) without perspective cues; almost all of the tested neurons (28/29) showed surface orientation selectivity for the SFS and/or the RDS without perspective cues. Eight of these 28 neurons (28.6%) showed selectivity for both the RDS and SFS, 7 (25.0%) were dominantly selective for the RDS, and 13 (46.4%) were dominantly selective for the SFS. These results suggest that neurons that show surface orientation tuning for the RDS without perspective cues compute surface orientation from the gradient of the binocular disparity given by the random-dot across the surface. On the other hand, neurons that show surface orientation tuning for the SFS without perspective cues may represent surface orientation primarily from the gradient of the binocular disparity along the contours. In conclusion, the SOS neurons in the area CIP are likely to operate higher order processing of disparity signals for surface perception by integrating the input signals from many disparity-sensitive neurons with different disparity tuning.

One of the major distinguishing features of the Dynamic Multiobjective Optimization Problems (DMOPs) is that optimization objectives will change over time, thus tracking the varying Pareto-Optimal Front (POF) becomes a challenge. One of the promising solutions is reusing "experiences" to construct a prediction model via statistical machine learning approaches. However, most existing methods neglect the non-independent and identically distributed nature of data to construct the prediction model. In this paper, we propose an algorithmic framework, called Tr-DMOEA, which integrates transfer learning and population-based evolutionary algorithms (EAs) to solve the DMOPs. This approach exploits the transfer learning technique as a tool to generate an effective initial population pool via reusing past experience to speed up the evolutionary process, and at the same time any population based multiobjective algorithms can benefit from this integration without any extensive modifications. To verify this idea, we incorporate the proposed approach into the development of three well-known evolutionary algorithms, nondominated sorting genetic algorithm II (NSGA-II), multiojective particle swarm 1 arXiv:1612.06093v2 [cs.NE] 18 Nov 2017 optimization (MOPSO), and the regularity model-based multiobjective estimation of distribution algorithm (RM-MEDA). We employ twelve benchmark functions to test these algorithms as well as compare them with some chosen state-of-the-art designs. The experimental results confirm the effectiveness of the proposed design for DMOPs.

We investigated the effects of linear perspective and binocular disparity, as monocular and binocular depth cues, respectively, on the response of surface-orientation-selective (SOS) neurons in the caudal part of the lateral bank of the intraparietal sulcus (area CIP). During the single-unit recording, monkeys were required to perform the delayed-matching-to-sample (successive same/different discrimination) of discriminating surface orientation in stereoscopic computer graphics. Of 211 visually responsive neurons, 66 were intensively tested using the solid-figure stereogram (SFS) of a square plate with both disparity and perspective cues (D+P condition), and 62 of these were identified as SOS neurons for responding selectively to the orientation of stimuli. All these neurons were further tested using a solid figure with perspective cues alone (P-only condition), and 58% (36/62) of these showed selective response to the orientation of the stimuli. Of the 62 SOS neurons, 35 neurons were also tested using SFS with disparity cues alone (D-only condition) in addition to the D+P and P-only conditions. We classified these 35 neurons into four groups by comparing the response selectivity under the P-only and D-only conditions. More than one-half of these (19/35) were sensitive to both perspective and disparity cues (DP neurons), and nearly one-third (11/35) of these were sensitive to disparity cues alone (D neurons), but a few (2/35) were sensitive to perspective cues alone (P neurons). The remaining (3/35) neurons exhibited orientation selectivity only when both cues were present. In DP neurons, the preferred orientation under the D+P condition was correlated to those under the D-only and P-only conditions, and the response magnitude under the D+P condition was greater than those under the D-only and P-only conditions, suggesting the integration of both cues for the perception of surface orientation. However, in these neurons, the orientation tuning sharpness under the D+P and D-only conditions was higher than that under the P-only condition, suggesting the dominance of disparity cues. After the single-unit recording experiments, muscimol was microinjected into the recording site to temporarily inactivate its function. In all three effective cases out of six microinjection experiments, discrimination of a three-dimensional (3D) surface orientation was impaired when disparity cues alone were present. In only one effective case, when a relatively large amount of muscimol was microinjected, discrimination of a 3D surface orientation was impaired even when both disparity and perspective cues were present. These results suggest that linear perspective is an important cue for representations of a 3D surface of SOS neurons in area CIP, although it is less effective than binocular disparity, and that both of these depth cues may be integrated in area CIP for the perception of surface orientation in depth.

No abstract

Domain adaptation learning (DAL) investigates how to perform a task across different domains. In this paper, we present a kernelized local-global approach to solve domain adaptation problems. The basic idea of the proposed method is to consider the global and local information regarding the domains (e.g., maximum mean discrepancy and intraclass distance) and to convert the domain adaptation problem into a bi-object optimization problem via the kernel method. A solution for the optimization problem will help us identify a latent space in which the distributions of the different domains will be close to each other in the global sense, and the local properties of the labeled source samples will be preserved. Therefore, classic classification algorithms can be used to recognize unlabeled target domain data, which has a significant difference on the source samples. Based on the analysis, we validate the proposed algorithm using four different sources of data: synthetic, textual, object, and facial image. The experimental results indicate that the proposed method provides a reasonable means to improve DAL algorithms.