Changes in visual processing strategy: effects of context, task, and training

我研究我们的大脑如何使用视觉信息，以便我们可以感知外部世界。为了做到这一点，我进行了心理物理实验，让人类参与者对刺激做出判断。我设计这些实验是为了将表现与视觉处理模型的预测进行比较。然后我可以研究大脑用来解决这些任务的策略。在接下来的研究中，我将探索这些处理策略在不同条件下如何变化。根据信号检测理论，视觉系统对噪声和不稳定的输入信号进行连续的统计推断。这将视觉的“问题”框定在不确定性方面。噪声输入用于将这种不确定性分解为感知体验。有人建议，该处理纳入了有效编码的原则。这就预示着应该有一种适应性策略，根据任务来调整它的处理过程，以前的证据表明视觉系统以这种方式适应，这既来自我自己的工作，也来自其他实验室。例如，我将研究一种可能的稳态平衡在整合噪声信号。更高的集成度降低了噪声，但代价是精度降低。通过测量行为，可以推断出潜在的处理过程。然而，可以设计所有针对单个底层函数的多个任务。一个例子是视觉轮廓连接成一个连续的概念的机制。最近，我开发了一种新的方法来探索这一点。从新任务中得出的结论与以前发现的不同。这可能是由于系统基于任务调整其轮廓处理。或者，刺激设计的差异可能会改变任务不可知的系统中的处理。我将使用不同的任务来限制对轮廓整合的理解，以及它与类似刺激下执行的其他功能的关系。 我感兴趣的是这些战略变化发生的时间尺度。它是否会根据预期的需求而先发制人地改变？还是必须在任务体验期间携带？还有一个问题是，效果如何在不同的任务之间转移。最近的研究表明，训练可以对视觉信息的处理方式产生根本性的影响。我将探索这一点。心理物理学允许我们直接通过视觉对行为的影响来研究视觉处理。我对人类如何执行视觉任务的研究将改进衡量这种表现的技术。了解大脑如何应对视觉所涉及的不断变化的挑战，也可能有助于我们了解它如何执行其他功能。这是计算机视觉研究人员感兴趣的，以揭示固定解决方案不足的情况。最后，了解我们能力的优势和劣势，可以让我们在设计人类交互系统时考虑到这些能力。

I investigate how our brains use visual information so that we may perceive the outside world. To do so, I perform psychophysical experiments where human participants make judgements about stimuli. I design these experiments to allow performance to be compared against predictions from models of visual processing. I can then investigate the strategies that the brain uses to solve these tasks. In my upcoming research, I will explore how these processing strategies change under various conditions. Under Signal Detection Theory, the visual system is performing a continuous statistical inference on noisy and unstable input signals. This frames the "problem" of vision in terms of uncertainty. The noisy input is used to resolve this uncertainty into a perceptual experience. It is proposed that the processing incorporates the principles of efficient coding. This would predict there should be an adaptive strategy, tailoring its processing based on the task posed to it. There is previous evidence that the visual system adapts in this way both from my own work and from other laboratories. For example, I will investigate a possible homeostatic balance in the integration of noisy signals. Greater integration reduces noise, but comes at the cost of a loss of precision. By measuring behaviour, inferences can be made about the underlying processing. It is possible however to design many tasks that all target a single underlying function. One example is the mechanism by which visual contours are joined-up into a continuous percept. Recently, I have developed a new method to probe this. The conclusions drawn from the new task differ from those found previously. This may be due to the system adjusting its contour processing on the basis of the task. Alternatively, differences in stimulus design may alter processing in a system which is task-agnostic. I will use different tasks to constrain an understanding of contour integration, and its relationship to other functions performed on similar stimuli. I am interested in the time scale over which these changes in strategy occur. Does it shift pre-emptively in response to the expected demands? Or must it be entrained during task experience? There is also the question of how effects would transfer between different tasks. Recent studies have shown that training can have a fundamental impact on how visual information is processed. I will explore this. Psychophysics allows us to investigate visual processing directly through is effects on behaviour. My research into how humans perform visual tasks will improve techniques to measure that performance. Understanding how the brain tackles the shifting challenges involved in vision may also help us see how it performs other functions. This is of interest to computer vision researchers, to reveal cases where fixed solutions are insufficient. Finally, knowing the strengths and weaknesses of our abilities allows them to be considered when designing systems for human interaction.