神经网络的终身学习机制与挑战：一项深度综述

需积分: 27 198 浏览量更新于2024-07-09 收藏 1.48MB PDF 举报

本文是一篇关于"神经网络的持续终身学习"的综述论文，由German I. Parisi等人撰写，发表于《神经网络》(Neural Networks)杂志，2019年卷113，页码54-71。论文背景指出，人类和高级动物在生命的各个阶段都能不断学习、微调和迁移知识与技能，这一现象被称为终身学习。终身学习对于计算机系统和人工智能至关重要，因为它决定了它们能否在现实世界中进行有效的交互，并处理连续变化的信息。该论文的核心内容涵盖了以下几个关键知识点： 1. **定义与概念**：终身学习被定义为一个复杂的过程，它涉及一系列神经认知机制，如感官运动技能的发展，以及长期记忆的巩固和检索。这强调了大脑的动态适应性，使得个体能够应对新经验而不至于忘记旧知识。 2. **挑战与问题**：文章讨论了在计算机科学领域，特别是神经网络中实现终身学习所面临的挑战，比如“灾难性遗忘”（catastrophic forgetting）。这是指在学习新任务时，模型可能会丧失之前学习的能力，这是由于权重调整导致的对旧信息的抑制。 3. **方法与策略**：为了克服灾难性遗忘，研究者提出了多种策略，包括但不限于：使用记忆增强技术（如记忆痕迹增强、记忆巩固）、增量学习（分阶段学习新任务，保持对旧任务的知识更新）、生成对抗网络（GANs）的应用，以及通过动态调整网络结构来适应不同阶段的学习需求。 4. **发展系统与模型**：论文探讨了模仿生物发展系统的模型，如发育阶段模拟、自组织学习等，以更好地理解和发展神经网络的终身学习能力。这些模型试图在不同阶段模仿人类和动物的学习过程，从而实现持续的知识扩展。 5. **应用前景**：作者认为，随着人工智能的发展，终身学习在自动驾驶、机器人技术、医疗诊断、自然语言处理等领域的潜力巨大。通过改进的神经网络架构和算法，这些应用将更加灵活和高效。 6. **未来方向**：论文结尾部分概述了未来研究可能关注的方向，如更深入地理解记忆机制、开发更有效的学习算法、以及集成跨学科知识，以提升人工智能的持续学习能力。这篇论文深入剖析了神经网络在实现终身学习中的理论与实践，以及面临的挑战和潜在解决方案，为推进人工智能在实际场景中的长期学习提供了有价值的参考。

G.I. Parisi, R. Kemker, J.L. Part et al. / Neural Networks 113 (2019) 54–71 57

deprivation of visual input began prior to ten weeks of age while no

changes were observed in adult animals. Additional experiments

showed that neural patterns of cortical organization can be driven

by external environmental factors at least for a period early in

development (Hubel & Wiesel, 1962, 1970; Hubel, Wiesel, & LeVay,

1977).

The most well-known theory describing the mechanisms of

synaptic plasticity for the adaptation of neurons to external stimuli

was first proposed by Hebb (1949), postulating that when one neu-

ron drives the activity of another neuron, the connection between

them is strengthened. More specifically, the Hebb’s rule states that

the repeated and persistent stimulation of the postsynaptic cell

from the presynaptic cell leads to an increased synaptic efficacy.

Throughout the process of development, neural systems stabilize

to shape optimal functional patterns of neural connectivity. The

simplest form of Hebbian plasticity considers a synaptic strength

w which is updated by the product of a pre-synaptic activity x and

the post-synaptic activity y:

∆w = x · y · η, (1)

where η is a given learning rate. However, Hebbian plasticity alone

is unstable and leads to runaway neural activity, thus requiring

compensatory mechanisms to stabilize the learning process (Ab-

bott & Nelson, 2000; Bienenstock, Cooper, & Munro, 1982). Stability

in Hebbian systems is typically achieved by augmenting Hebbian

plasticity with additional constraints such as upper limits on the

individual synaptic weights or average neural activity (Miller &

MacKay, 1994; Song, Miller, & Abbott, 2000). Homeostatic mech-

anisms of plasticity include synaptic scaling and meta-plasticity

which directly affect synaptic strengths (Davis, 2006; Turrigiano,

2011). Without loss of generality, homeostatic plasticity can be

viewed as a modulatory effect or feedback control signal that

regulates the unstable dynamics of Hebbian plasticity (see Fig. 1a).

The feedback controller directly affects synaptic strength on the

basis of the observed neural activity and must be fast in relation to

the timescale of the unstable system (Aström & Murray, 2010). In

its simplest form, modulated Hebbian plasticity can be modelled

by introducing an additional modulatory signal m to Eq. (1) such

that the synaptic update is given by

∆w = m · x · y · η. (2)

Modulatory feedback in Hebbian neural networks has received

increasing attention, with different approaches proposing biologi-

cally plausible learning through modulatory loops (Grant, Tanner,

& Itti, 2017; Soltoggio et al., 2017). For a critical review of the

temporal aspects of Hebbian and homeostatic plasticity, we refer

the reader to Zenke, Gerstner et al. (2017).

Evidence on cortical function has shown that neural activity in

multiple brain areas results from the combination of bottom-up

sensory drive, top-down feedback, and prior knowledge and ex-

pectations (Heeger, 2017). In this setting, complex neurodynamic

behaviour can emerge from the dense interaction of hierarchically

arranged neural circuits in a self-organized manner (Tani, 2016).

Input-driven self-organization plays a crucial role in the brain

(Nelson, 2000), with topographic maps being a common feature

of the cortex for processing sensory input (Willshaw & von der

Malsburg, 1976). Different models of neural self-organization have

been proposed that resemble the dynamics of basic biological

findings on Hebbian-like learning and plasticity (Fritzke, 1992;

Kohonen, 1982; Marsland, Shapiro, & Nehmzow, 2002; Martinetz,

Berkovich, & Schulten, 1993), demonstrating that neural map or-

ganization results from unsupervised, statistical learning with

nonlinear approximations of the input distribution. To stabilize the

unsupervised learning process, neural network self-organization

can be complemented with top-down feedback such as task-

relevant signals that modulate the intrinsic map plasticity (Parisi

et al., 2018; Soltoggio et al., 2017). In a hierarchical processing

regime, neural detectors have increasingly large spatio-temporal

receptive fields to encode information over larger spatial and tem-

poral scales (Hasson, Yang, Vallines, Heeger, & Rubin, 2008; Taylor,

Hobbs, Burroni, & Siegelmann, 2015). Thus, higher-level layers

can provide the top-down context for modulating the bottom-

up sensory drive in lower-level layers. For instance, bottom-up

processing is responsible for encoding the co-occurrence statis-

tics of the environment while error-driven signals modulate this

feedforward process according to top-down, task-specific factors

(Murray et al., 2016). Together, these models contribute to a better

understanding of the underlying neural mechanisms for the devel-

opment of hierarchical cortical organization.

2.3. The complementary learning systems

The brain learns and memorizes. The former task is charac-

terized by the extraction of the statistical structure of the per-

ceived events with the aim to generalize to novel situations. The

latter, conversely, requires the collection of separated episodic-

like events. Consequently, the brain must comprise a mecha-

nism to concurrently generalize across experiences while retaining

episodic memories.

Sophisticated cognitive functions rely on canonical neural cir-

cuits replicated across multiple areas (Douglas, Koch, Mahowald,

Martin, & Suarez, 1995). However, although there are shared struc-

tural properties, different brain areas operate at multiple timescales

and learning rates, thus differing significantly from each other

in a functional way (Benna & Fusi, 2016; Fusi, Drew, & Abbott,

2005). A prominent example is the complementary contribution

of the neocortex and the hippocampus in learning and memory

consolidation (McClelland et al., 1995; O’Reilly, 2004; O’Reilly &

Norman, 2002). The complementary learning systems (CLS) the-

ory (McClelland et al., 1995) holds that the hippocampal system

exhibits short-term adaptation and allows for the rapid learning

of novel information which will, in turn, be played back over

time to the neocortical system for its long-term retention (see

Fig. 1b). More specifically, the hippocampus employs a rapid learn-

ing rate and encodes sparse representations of events to mini-

mize interference. Conversely, the neocortex is characterized by

a slow learning rate and builds overlapping representations of

the learned knowledge. Therefore, the interplay of hippocampal

and neocortical functionality is crucial to concurrently learn reg-

ularities (statistics of the environment) and specifics (episodic

memories). Both brain areas are known to learn via Hebbian and

error-driven mechanisms (O’Reilly & Rudy, 2000). In the neocortex,

feedback signals will yield task-relevant representations while, in

the case of the hippocampus, error-driven modulation can switch

its functionally between pattern discrimination and completion for

recalling information (O’Reilly, 2004).

Studies show that adult neurogenesis contributes to the forma-

tion of new memories (Altman, 1963; Cameron, Woolley, McEwen,

& Gould, 1993; Eriksson et al., 1998; Gage, 2000). It has been

debated whether human adults grow significant amounts of new

neurons. Recent research has suggested that hippocampal neuro-

genesis drops sharply in children to undetectable levels in adult-

hood (Sorrells et al., 2018). On the other hand, other studies suggest

that hippocampal neurogenesis sustains human-specific cogni-

tive function throughout life (Boldrini, Fulmore, Tartt, Simeon, &

Pavlova, 2018). During neurogenesis, the hippocampus’ dentate

gyrus uses new neural units to quickly assimilate and immediately

recall new information (Altman, 1963; Eriksson et al., 1998). Dur-

ing initial memory formation, the new neural progenitor cells ex-

hibit high levels of plasticity; and as time progresses, the plasticity

decreases to make the new memory more stable (Deng, Aimone,

& Gage, 2010). In addition to neurogenesis, neurophysiological

剩余17页未读，继续阅读

syp_net

粉丝: 159
资源: 1187

神经网络的终身学习机制与挑战：一项深度综述

神经网络中的元学习综述

论文研究-基于终身学习RBF神经网络的网络安全态势预测 .pdf

翻译：神经网络语言模型论文综述_ lm综述.zip

图神经网络综合综述学习第一周

awesome-lifelong-continual-learning:终身持续机器学习领域的论文，博客，数据集和软件列表

深度学习驱动的卷积神经网络进展与应用综述

深度学习驱动的卷积神经网络发展与应用综述

图机器学习综述：网络嵌入、图正则化神经网络和图神经网络

多模态机器学习综述论文PPT深度分享

图神经网络自监督学习：一份综合综述

最新资源