“Here are 2 good papers--one from last year, one very recent--that might change your mind.”
One-shot Learning with Memory-Augmented Neural Networks
用外部记忆让网络看几个样本就能学会新类别
单样本学习对深度网络仍是难题:传统基于梯度的网络需要大量数据和反复迭代,遇到新数据时还要低效地重学参数并易发生灾难性干扰。论文展示,带外部记忆的记忆增强神经网络(如神经图灵机 NTM)能快速吸收新数据,并在只见到少量样本后做出准确预测;作者还提出一种只基于内容、不再使用位置聚焦机制来访问外部记忆的新方法。
2016 年的评论者用它反驳「神经网络需要海量数据」的观点,作为网络能从少量样本学习的新进展,也有人指出相关方法扩展性不佳,并提到 DeepMind 当时的最好结果。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“Newer work is even able to learn from just a few examples[0] and then generalize beyond that.”
“This is just an outdated view of the state-of-the-art. It's understandable, given that it's outdated by maybe six months, iff you're willing to go with preprints.”
“A paper someone pointed me to recently on HN suggested that it doesn't scale very well:”
“And yet some great research is being done that will prove him wrong on his first point.”
“I think the SOTA is probably [2], which came out of DeepMind.”
看全部 7 条 →
“including: [LINK], One-shot Learning with Memory-Augmented Neural Networks and [OTHER_PAPER].”
- One-shot Learning with Memory-Augmented Neural Networks ↗astdb · 2016-05-20 · 103 分