DAVID H. ACKLEY GEOFFREY E. HINTON 卡内基梅隆大学计算机科学系 TERRENCE J. SEJNOWSKI 约翰霍普金斯大学生物物理学系 由简单处理单元组成的大规模并行网络的计算能力，在于元件间硬件连接所提供的通信带宽。这些连接能够使系统知识的很大一部分在极短时间内应用于问题实例。大规模并行网络似乎特别适合的一种计算是大型约束满足搜索，但要高效利用这些连接，必须满足两个条件：首先，必须找到一种适合并行网络的搜索技术。其次，必须有一种选择内部表示的方法，使得预先存在的硬件连接能够被高效地用于编码搜索领域中的约束。我们描述了一种基于统计力学的一般并行搜索方法，并展示了它如何导出一个一般学习规则，用于修改连接强度，从而以高效的方式融入关于任务领域的知识。我们描述了一些简单例子，其中学习算法创建的内部表示被证明是使用预先存在的连接结构的最有效方式。 关于大脑结构和新 VLSI 技术潜力的证据，引发了对“连接主义”系统兴趣的复兴。
DAVID H. ACKLEY GEOFFREY E. HINTON Computer Science Department Carnegie-Mellon University TERRENCE J. SEJNOWSKI Biophysics Department The Johns Hopkins University The computational power of massively parallel networks of simple processing elements resides in the communication bandwidth provided by the hardware connections between elements. These connections can allow a significant fraction of the knowledge of the system to be applied to an instance of a problem in a very short time. One kind of computation for which massively parallel networks appear to be well suited is large constraint satisfaction searches, but to use the connections efficiently two conditions must be met: First, a search technique that is suitable for parallel networks must be found. Second, there must be some way of choosing internal representations which allow the preexisting hardware connections to be used efficiently for encoding the constraints in the domain being searched. We describe a general parallel search method, based on statistical mechanics, and we show how it leads to a general learning rule for modifying the connection strengths so as to incorporate knowledge about a task domain in an efficient
核心贡献 · Key contributions
提出玻尔兹曼机，一种基于统计力学的并行随机搜索方法，用于约束满足。 Proposed the Boltzmann Machine, a parallel stochastic search method for constraint satisfaction based on statistical mechanics.
推导了一个局部学习规则，通过调整连接强度来建模环境样本的概率分布。 Derived a local learning rule that adjusts connection strengths to model the probability distribution of environmental examples.
展示了隐藏单元能自动形成内部表示，例如在编码器问题中发现二进制码。 Demonstrated that hidden units can automatically form internal representations, such as binary codes in encoder problems.
表明退火和噪声注入有助于逃离局部极小值，改善全局优化。 Showed that annealing and noise injection enable escape from local minima, improving global optimization.
将统计力学与并行计算联系起来，影响了后续的可解释性和学习模型。 Connected statistical mechanics to parallel computation, influencing later interpretability and learning models.
局限 · Limitations
由于需要大量模拟退火和均衡采样，学习速度极其缓慢。 Learning is extremely slow due to the need for many simulated annealings and equilibrium sampling.
需要仔细调整参数（如温度调度、噪声水平）才能实现稳定学习。 Careful tuning of parameters (e.g., temperature schedule, noise level) is required for stable learning.
假设二元单元和对称权重，限制了在实值或非对称系统中的应用。 Assumes binary units and symmetric weights, limiting applicability to real-valued or asymmetric systems.
当隐藏单元产生复杂误差曲面时，容易陷入局部极小值。 Easily trapped in local minima when hidden units create complex error landscapes.
由于对称连接和无传输延迟的假设，在生物学上不合理。 Biologically implausible due to assumptions of symmetric connections and no transmission delays.