CRCNS:Proto-object based perceptual organization in three dimensions

CRCNS：基于原型对象的三维感知组织

• 批准号: 9336905
• 负责人: Ralph Etienne-Cummings
• 金额: $ 37.44万
• 依托单位: JOHNS HOPKINS UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-01 至 2019-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9336905
• 关键词: Address; Agreement; Area; Attention; Biological; Biological Sciences; Biotechnology; Brain; CCL7 gene; Cells; Code; Cognition; Cognitive; Collaborations; Communication Research; Computer Simulation; Development; Dimensions; Educational workshop; Elements; Engineering; Faculty; Feedback; Fire - disasters; Goals; Grouping; Human; Image; Income; Individual; Interdisciplinary Study; Lead; Link; Mathematics; Mentors; Minority-Serving Institution; Modeling; Neural Network Simulation; Neurons; Participant; Pattern; Perception; Population; Primates; Recording of previous events; Recurrence; Research; Research Ethics; Research Project Grants; Retinal; Role; Rotation; STEM research; Science; Sensory; Shapes; Signal Transduction; Site; Speed; Structure; Student recruitment; Students; System; Technology; Testing; Theoretical Studies; Time; Training; Visual; Visual system structure; Work; awake; base; cognitive process; community college; experimental study; graduate student; high school; insight; member; models and simulation; neuronal circuitry; nonhuman primate; object recognition; perceptual organization; programs; relating to nervous system; response; selective attention; simulation; skills; student mentoring; success; tool; two-dimensional; undergraduate student; visual information

The visual brain infers a three-dimensional world from twodimensional images and organizes the visual information in terms of objects in three-dimensional space, representing even objects that are partially occluded and appear fragmented in the retinal image. This organization is the basis for attentive selection, action planning and object recognition. A combination of experimental and theoretical studies together with model implementations in neuromorphic hardware will be used to elucidate the interface between visual feature representations and attentive cognitive processes. Previous findings on the neural coding of figure-ground structure can be understood in terms of grouping mechanisms that structure the incoming sensory information as proto-objects (objects as defined by the system at this stage). The grouping mechanisms also provide handles for top-down mechanisms to address and select object-related information. The proposed work will explain how neuronal circuitry organizes spatially disconnected visual features into perceptual objects. How is this implemented neurally to lead to a coherent representation? Detailed computational models of the underlying circuitry will be developed, both as standard numerical simulations and in fast, neuromorphic hardware, and then tested by multiple single-cell recordings in awake non-human primates. Specifically, while prior studies examined spike time correlations indiscriminately in all neurons, our recent studies differentiated neurons according to their role in the grouping circuits. The grouping hypothesis predicts elevated synchrony only in pairs of neurons that belong to the same grouping circuit, but not in other pairs. These model predictions were confirmed in a recent study which showed that spike-spike correlation functions are in qualitative agreement with the idea that perceptual grouping is implemented by feedback from populations of dedicated grouping cells. Quantitative understanding requires the development of explicit spiking models, which is one of the main foci of this proposal. Models will be implemented on neuromorphic spiking hardware since the complexity of the cortical circuitry makes realistic model simulations on CPU/GPU system impossible. Predictions of integrate-and-fire type models of this circuitry will be compared with rate and synchrony observed in our recordings and deviates used to fine-tune the models. We will pursue the educational and broader impacts aims on five fronts. 1) Students will be crosstrained and mentored in biological, mathematical and engineering sciences, which will lead to graduates with unique skill sets. 2) We will contribute to the development of the nascent neuromorphic engineering field, providing new research problems that can benefit from the crosstraining and collaboration. We plan to participate in the NSF sponsored Telluride Neuromorphic Cognition Engineering and Capo Caccia Neuromorphic Cognitive Engineering Workshops for this purpose. 3)We will provide an opportunity for undergraduate students to participate in the research as part of our Site REU (managed by one of the PIs). They are trained in communications, research ethics and project management, which are crucial for success in todays biotechnology and bioscience work and market place. 4) We currently host students from local high-schools who conduct STEM research practicum rotations in our labs. This project will provide a perfect venue for the rotators to get exposed and mentored on multi-disciplinary research problems. We will use a tiered mentoring structure, where undergraduates mentor K-12 rotators, graduate students mentor undergraduates, and faculty members mentor all participants. 5) Our student recruitment plans will build on our current partnerships with MARC, LSAMP, McNair, SWE, SHPE and other similar programs and minority-serving institutions and local community colleges, to help develop a pipeline of qualified, diverse individuals who will contribute to the workforce in the area of STEM.

视觉大脑从二维推断出三维世界 图像并根据三维物体组织视觉信息 空间，甚至代表部分被遮挡并在视网膜中呈现碎片的物体 图像。这种组织是细心选择、行动计划和对象识别的基础。一个 实验和理论研究与模型实施相结合 神经形态硬件将用于阐明视觉特征表示之间的接口 和专注的认知过程。先前关于图形-背景结构的神经编码的发现可以 可以根据分组机制来理解，该机制将传入的感官信息构造为 原型对象（系统在此阶段定义的对象）。分组机制还提供 用于处理和选择对象相关信息的自上而下机制的句柄。拟议的工作 将解释神经回路如何将空间上断开的视觉特征组织成感知 对象。如何在神经上实现这一点以产生连贯的表示？详细计算 将开发底层电路的模型，既作为标准数值模拟，又作为快速、 神经形态硬件，然后在清醒的非人类身上通过多个单细胞记录进行测试 灵长类动物。具体来说，虽然先前的研究不加区别地检查了所有情况下的尖峰时间相关性 神经元，我们最近的研究根据神经元在分组电路中的作用来区分神经元。这 分组假设预测仅在属于同一神经元对中同步性提高 分组电路，但不在其他对中。这些模型预测在最近的一项研究中得到了证实，该研究 表明尖峰-尖峰相关函数与感知的观点在定性上一致 分组是通过来自专用分组单元群体的反馈来实现的。定量 理解需要开发明确的尖峰模型，这是该研究的主要焦点之一 提议。由于神经形态尖峰硬件的复杂性，模型将在神经形态尖峰硬件上实现 皮质电路使得 CPU/GPU 系统上的真实模型模拟变得不可能。的预测 该电路的集成和发射型模型将与我们在我们的系统中观察到的速率和同步性进行比较 用于微调模型的记录和偏差。 我们将在五个方面追求教育和更广泛影响的目标。 1）学生将接受交叉训练 并在生物、数学和工程科学方面接受指导，这将导致 具有独特技能的毕业生。 2）我们将为新生神经形态的发展做出贡献 工程领域，提供新的研究问题，可以从交叉训练和 合作。我们计划参加 NSF 赞助的特柳赖德神经形态认知 为此目的举办了工程和 Capo Caccia 神经形态认知工程研讨会。 3）我们 将为本科生提供参与我们网站的研究的机会 REU（由一名 PI 管理）。他们接受过沟通、研究伦理和项目方面的培训 管理，这对于当今生物技术和生物科学工作和市场的成功至关重要 地方。 4) 我们目前接待来自当地高中的学生进行STEM研究实习 在我们的实验室轮换。该项目将为旋转者提供一个完美的暴露和展示场所 指导多学科研究问题。我们将使用分层指导结构，其中 本科生指导 K-12 轮转人员，研究生指导本科生和教职员工 指导所有参与者。 5) 我们的学生招生计划将建立在我们目前与 MARC、LSAMP、McNair、SWE、SHPE 和其他类似计划以及少数族裔服务机构和 当地社区大学，帮助培养合格的、多样化的人才，他们将做出贡献 STEM 领域的劳动力。