2.24 考研英语阅读模拟试题【难度7.8】

Section II Reading Comprehension 

Part A 

Directions: Read the following four texts. Answer the questions below each text by choosing A, B, C or D. Mark your answers on the ANSWER SHEET. 

Text 4

Autonomous vehicles face many hazards as they set out on the road. Cyclists swerve in and out of traffic, distracted pedestrians amble into the street, and human drivers change lanes without indicating. Accurate vision and quick reflexes are required. Until now, even the best robots struggled to make sense of such complex environments as quickly as humans. But in a study published in Nature Communications, researchers took inspiration from the human eye to develop a new artificial-vision system that is four times faster than the current state of the art.

Many robots are equipped with cameras to see the world. A popular approach to interpret these images as motion is optical flow. As a robot moves, optical-flow algorithms convert shifting patterns of brightness into information about its own movement and surrounding objects. These algorithms allow robots to safely navigate streets or perform precision surgery.

However, optical-flow methods are computationally intensive because every pixel in each frame must be processed. Even with state-of-the-art technology, distinguishing objects in a single frame can take over 0.6 seconds. For an autonomous vehicle driving at motorway speeds, every half-second delay leads to around 12 metres of travel with outdated information. If artificial systems are to safely navigate dynamic environments, their eyesight needs an upgrade.

Shuo Gao, a roboticist at Beihang University, wondered if biology might have the answer. Human eyes tame the complexity of the world by focusing attention only where it is needed. Central to this process is a region of the brain known as the lateral geniculate nucleus (LGN).  The LGN acts as a relay station in the visual pathway, receiving information from the retina and passing it to the brain’s visual cortex. Crucially, the LGN plays a filtering role, indicating to the visual cortex where processing power should be prioritised. Because its filter is sensitive to changes in time and space, it allows the brain to efficiently track rapid movement.

Dr. Gao and his team aimed to introduce an LGN-like layer into artificial vision systems to guide the attention of optical-flow algorithms. Doing so with traditional computer chips—where circuits that process information are kept separate from those that store data—would not have given them the necessary speed. Instead, they turned to neuromorphic hardware, which mimics the human brain by integrating processing and storage functions into the same circuitry.

The researchers developed a novel neuromorphic kit to imitate the LGN. By tracking light intensity changes over time, the device builds a picture of where motion is occurring and prioritises those regions for optical-flow analysis.

Testing the new setup in contexts like autonomous driving, scientists found the system operated at four times the speed of existing methods while maintaining or improving accuracy. For autonomous driving, accuracy actually doubled.

The system is not without limitations. The neuromorphic hardware must still feed information back to conventional algorithms; it can never completely overcome those algorithms’ inherent shortcomings. Indeed, accuracy decreased for scenes with complex, dense motion—a familiar hurdle for optical flow. Nevertheless, the researchers hope their system will increase the complexity of scenarios in which robotics can be safely deployed, helping human and machine see eye to eye.

36. According to Paragraph 3, the major drawback of current optical-flow methods is that they _______.

[A] fail to recognize stationary objects on motorways 

[B] demand excessive computing power and cause dangerous delays 

[C] rely too heavily on outdated satellite navigation information 

[D] process only a small fraction of pixels in a given frame

37. In the human visual system, the lateral geniculate nucleus (LGN) functions primarily to _______.

[A] convert visual stimuli into electrical signals directly 

[B] store visual memory for the brain's visual cortex 

[C] filter visual information and prioritize processing resources 

[D] process the exact shapes and colors of moving objects

38. Why did Dr. Gao's team choose neuromorphic hardware over traditional computer chips?

[A] Traditional chips consume too much physical space in robots. 

[B] Neuromorphic hardware separates data storage from processing. 

[C] Traditional chips lack the integrated architecture needed for faster speeds. 

[D] Neuromorphic hardware is significantly cheaper to mass-produce.

39. A limitation of the newly developed artificial-vision system is its _______.

[A] reliance on conventional algorithms when handling dense motion 

[B] inability to track light intensity changes over a long period 

[C] lower accuracy compared to older optical-flow methods 

[D] failure to surpass human-level reflex speeds in driving tests

40. Which of the following would be the best title for the text?

[A] The Hidden Dangers of Autonomous Vehicles 

[B] Neuromorphic Chips: The End of Traditional Computing 

[C] How the Human Eye Processes Optical Illusions 

[D] Taking a Cue from Biology to Upgrade Robotic Vision

附注：根据历年考研英语真题阅读题源外刊等，摘选最新文章，模拟仿真出题。

参考答案见以下。

Quick look: BCCAD

36. 【答案】B

【解析】题型：因果/细节题

定位： 第三段 "...computationally intensive because every pixel... must be processed... take over 0.6 seconds... leads to around 12 metres of travel with outdated information."

分析： 第三段明确指出，光流法的缺点是计算量大（computationally intensive），处理单帧需0.6秒，这种延迟（delay）在高速行驶时非常危险（带旧信息行驶12米）。这完全对应选项 B。

干扰项：[A] 文中主要讨论动态物体（motion），并未说无法识别静止物体。 [C] outdated information 指的是由于处理延迟导致的环境信息过时，并非“卫星导航”信息。 [D] 与原文相反，原文说“每个像素（every pixel）”都必须被处理，导致计算量大。

37. 【答案】C

【解析】题型：细节题

定位： 第四段 "...the LGN plays a filtering role, indicating to the visual cortex where processing power should be prioritised."

分析： 文中指出，LGN 在视觉通路中起到了“过滤（filtering）”作用，指示视觉皮层应该优先将“处理能力（processing power）”放在哪里。选项 C 完美概括了这一功能。

干扰项：[A] 将视觉刺激转化为电信号是“视网膜（retina）”的功能。 [B] 文中未提及 LGN 存储视觉记忆。 [D] LGN 主要是追踪快速移动并分配注意力，具体的形状颜色处理通常是视觉皮层的工作。

38. 【答案】C

【解析】题型：因果细节题

定位： 第五段 "Doing so with traditional computer chips—where circuits that process information are kept separate from those that store data—would not have given them the necessary speed. Instead, they turned to neuromorphic hardware... integrating processing and storage..."

分析： 传统芯片将处理电路和存储电路分开（kept separate），这无法提供所需的速度。神经形态硬件则将两者整合（integrating）。因此，选择神经形态硬件是因为传统芯片缺乏能带来更快速度的整合架构。

干扰项：[A] 文中讨论的是速度和架构，不是物理空间。 [B] 恰恰相反，神经形态硬件是“整合（integrating）”这两者，传统芯片才是“分离（separating）”的。 [D] 文中并未提及生产成本。

39. 【答案】A

【解析】题型：细节题

定位： 第八段 "The neuromorphic hardware must still feed information back to conventional algorithms... accuracy decreased for scenes with complex, dense motion..."

分析： 系统的局限性在于，它仍然必须将信息反馈给传统算法，无法克服传统算法的固有缺陷，在处理复杂、密集的运动场景时准确率会下降。选项 A 概括了这一依赖性导致的局限。

干扰项：[B] 第六段指出它能通过追踪光强度变化来工作，未说无法长时间追踪。 [C] 第七段指出其准确率是保持或提高（maintaining or improving accuracy），在自动驾驶中甚至翻倍。 [D] 第七段说明该系统在大多数情况下已经超越了人类速度（surpassed human-level speeds）。

40. 【答案】D

【解析】题型：标题主旨题

定位： 全文主旨线索。

分析： 文章开篇提出机器视觉的困境，接着指出研究人员从人类眼睛（生物学）中汲取灵感，利用模仿大脑的神经形态硬件改进了机器（机器人/自动驾驶）的视觉系统。选项 D “从生物学中汲取灵感以升级机器视觉”准确涵盖了全文核心。

干扰项：[A] 自动驾驶的危险只是第一段的引入背景。 [B] 夸大其词，文中并未宣称传统计算的终结。 [C] 探讨的是如何借鉴人眼，而不是研究人眼如何处理视错觉。

【参考译文】

自动驾驶汽车在上路时面临着许多危险。骑行者在车流中忽左忽右地穿梭，分心的行人漫不经心地走入车道，人类驾驶员变道时不打转向灯。这需要精确的视觉和快速的反应。直到现在，即使是最优秀的机器人，也很难像人类一样快速地理解这种复杂的环境。但在发表于《自然·通讯》的一项研究中，研究人员从人类眼睛中汲取灵感，开发出一种新的机器视觉系统，其速度是当前最先进技术的四倍。

许多机器人都配备了摄像头来观察世界。一种将这些图像解释为运动的流行方法是“光流法（optical flow）”。当机器人移动时，光流算法将亮度变化模式转换为有关其自身运动和周围物体的信息。这些算法让机器人能够安全地在街道上导航或进行精密手术。

然而，光流法是计算密集型的，因为必须处理每一帧中的每个像素。即使使用最先进的技术，在单帧中区分物体也可能需要 0.6 秒以上。对于以高速公路速度行驶的自动驾驶汽车来说，每半秒的延迟就会导致携带过时信息行驶约 12 米。如果人工智能系统要在动态环境中安全导航，它们的视力需要升级。

北京航空航天大学的机器人学家高硕想知道生物学是否能提供答案。人类的眼睛通过只把注意力集中在需要的地方，来驯服这个世界的复杂性。这一过程的核心是大脑中被称为“外侧膝状体核（LGN）”的区域。LGN 充当视觉通路中的中继站，接收来自视网膜的信息，并将其传递给大脑的视觉皮层。至关重要的是，LGN 发挥着过滤作用，向视觉皮层指示应优先将处理能力放在何处。因为它的过滤器对时间和空间的变化很敏感，所以它允许大脑有效地追踪快速移动。

高博士和他的团队旨在将类似 LGN 的过滤层引入机器视觉系统，以引导光流算法的注意力。如果使用传统的计算机芯片——处理信息的电路和存储数据的电路是分开的——并不能给他们带来所需的速度提升。相反，他们转向了神经形态硬件（neuromorphic hardware），这种硬件通过将处理和存储功能整合到同一电路中来模仿人类大脑。

研究人员开发了一种新型的神经形态套件来模仿 LGN。通过追踪光强度随时间的变化，该设备能够构建出特定环境中发生运动的画面，并优先对这些区域进行光流分析。

科学家们在自动驾驶等环境中测试了这种新设置，发现该系统在保持或提高准确率的同时，运行速度大约是现有光流法的四倍。在自动驾驶方面，准确率实际上翻了一倍。

该系统并非没有局限性。神经形态硬件仍然必须将信息反馈给传统算法；它永远无法完全克服那些算法的固有缺陷。事实上，研究人员观察到，在复杂、密集运动的场景中，准确率会下降——这是光流法常见的一个障碍。尽管如此，研究人员希望他们的新系统能增加机器人安全部署场景的复杂性，帮助人类和机器“目光一致”。

【词汇注释】

swerve: verb (CHANGE DIRECTION) to change direction suddenly, especially in order to avoid hitting somebody/something  急转弯；突然转向（文中指骑行者在车流中穿梭/猛拐）

amble: verb (WALK) to walk at a slow, relaxed speed 漫步；缓行（文中指分心的行人慢吞吞地走入车道）

computationally intensive: phrase (COMPUTING) requiring a large amount of computing power and resources to perform a task 计算密集的（文中指光流法需要极大的算力）

retina: noun (EYE) a layer of tissue at the back of the eye that is sensitive to light and sends signals to the brain 视网膜（将光刺激转化为电信号的器官）

neuromorphic: adjective (TECHNOLOGY) relating to computing systems that are inspired by the structure and function of the human brain 神经形态的（文中指整合了处理和存储功能的仿生芯片硬件）

dense: adjective (THICK) containing a lot of people, things, plants, etc. with little space between them 密集的；稠密的（文中指场景中存在大量且拥挤的运动物体）

附注：

 ©图源水印/网络