The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.synaptic consolidation | artificial intelligence | stability plasticity | continual learning | deep learning A chieving artificial general intelligence requires that agents are able to learn and remember many different tasks (1). This is particularly difficult in real-world settings: The sequence of tasks may not be explicitly labeled, tasks may switch unpredictably, and any individual task may not recur for long time intervals. Critically, therefore, intelligent agents must demonstrate a capacity for continual learning: that is, the ability to learn consecutive tasks without forgetting how to perform previously trained tasks.Continual learning poses particular challenges for artificial neural networks due to the tendency for knowledge of the previously learned task(s) (e.g., task A) to be abruptly lost as information relevant to the current task (e.g., task B) is incorporated. This phenomenon, termed catastrophic forgetting (2-6), occurs specifically when the network is trained sequentially on multiple tasks because the weights in the network that are important for task A are changed to meet the objectives of task B. Whereas recent advances in machine learning and in particular deep neural networks have resulted in impressive gains in performance across a variety of domains (e.g., refs. 7 and 8), little progress has been made in achieving continual learning. Current approaches have typically ensured that data from all tasks are simultaneously available during training. By interleaving data from multiple tasks during learning, forgetting does not occur because the weights of the network can be jointly optimized for performance on all tasks. In this regime-often referred to as the multitask learning paradigm-deep-learning techniques have been used to train single agents that can successfully play multiple Atari games (9, 10). If tasks are presented sequentially, multitask learning can be used only if the data are recorded by an episodic memory system and replayed to the network during training. This approach [often called system-level consolidation (4, 5)] is impractical for learning large numbers of tasks, as in our setting it would require the amount of memories being stored and replayed to be proportional to the number of tasks. The lack of algorithms to support continual learning thus rema...

Theano is a compiler for mathematical expressions in Python that combines the convenience of NumPy's syntax with the speed of optimized native machine language. The user composes mathematical expressions in a high-level description that mimics NumPy's syntax and semantics, while being statically typed and functional (as opposed to imperative). These expressions allow Theano to provide symbolic differentiation. Before performing computation, Theano optimizes the choice of expressions, translates them into C++ (or CUDA for GPU), compiles them into dynamically loaded Python modules, all automatically. Common machine learning algorithms implemented with Theano are from 1.6× to 7.5× faster than competitive alternatives (including those implemented with C/C++, NumPy/SciPy and MATLAB) when compiled for the CPU and between 6.5× and 44× faster when compiled for the GPU. This paper illustrates how to use Theano, outlines the scope of the compiler, provides benchmarks on both CPU and GPU processors, and explains its overall design.

No abstract

Deep neural networks have achieved impressive successes in fields ranging from object recognition to complex games such as Go. Navigation, however, remains a substantial challenge for artificial agents, with deep neural networks trained by reinforcement learning failing to rival the proficiency of mammalian spatial behaviour, which is underpinned by grid cells in the entorhinal cortex . Grid cells are thought to provide a multi-scale periodic representation that functions as a metric for coding space and is critical for integrating self-motion (path integration) and planning direct trajectories to goals (vector-based navigation). Here we set out to leverage the computational functions of grid cells to develop a deep reinforcement learning agent with mammal-like navigational abilities. We first trained a recurrent network to perform path integration, leading to the emergence of representations resembling grid cells, as well as other entorhinal cell types . We then showed that this representation provided an effective basis for an agent to locate goals in challenging, unfamiliar, and changeable environments-optimizing the primary objective of navigation through deep reinforcement learning. The performance of agents endowed with grid-like representations surpassed that of an expert human and comparison agents, with the metric quantities necessary for vector-based navigation derived from grid-like units within the network. Furthermore, grid-like representations enabled agents to conduct shortcut behaviours reminiscent of those performed by mammals. Our findings show that emergent grid-like representations furnish agents with a Euclidean spatial metric and associated vector operations, providing a foundation for proficient navigation. As such, our results support neuroscientific theories that see grid cells as critical for vector-based navigation, demonstrating that the latter can be combined with path-based strategies to support navigation in challenging environments.

After a more than decade-long period of relatively little research activity in the area of recurrent neural networks, several new developments will be reviewed here that have allowed substantial progress both in understanding and in technical solutions towards more efficient training of recurrent networks. These advances have been motivated by and related to the optimization issues surrounding deep learning. Although recurrent networks are extremely powerful in what they can in principle represent in terms of modeling sequences, their training is plagued by two aspects of the same issue regarding the learning of long-term dependencies. Experiments reported here evaluate the use of clipping gradients, spanning longer time ranges with leaky integration, advanced momentum techniques, using more powerful output probability models, and encouraging sparser gradients to help symmetry breaking and credit assignment. The experiments are performed on text and music data and show off the combined effects of these techniques in generally improving both training and test error.

No abstract

Artificial intelligence research has seen enormous progress over the past few decades, but it predominantly relies on fixed datasets and stationary environments. Continual learning is an increasingly relevant area of study that asks how artificial systems might learn sequentially, as biological systems do, from a continuous stream of correlated data. In the present review, we relate continual learning to the learning dynamics of neural networks, highlighting the potential it has to considerably improve data efficiency. We further consider the many new biologically inspired approaches that have emerged in recent years, focusing on those that utilize regularization, modularity, memory, and meta-learning, and highlight some of the most promising and impactful directions. The World Is Not Stationary A common benchmark for success in artificial intelligence is the ability to emulate human learning. We measure the abilities of humans to recognize images, play games, and drive a car, to name a few, and then develop machine learning models that can match or exceed these given enough training data. This paradigm puts the emphasis on the end result, rather than the learning process, and overlooks a critical characteristic of human learning: that it is robust to changing tasks and sequential experience. It is perhaps unsurprising that humans can learn this way, after all, time is irreversible and the world is non-stationary (see Glossary), so human learning has evolved to thrive in dynamic learning settings. However, this robustness is in stark contrast to the most powerful modern machine learning methods, which perform well only when presented with data that are carefully shuffled, balanced, and homogenized. Not only do these models underperform when presented with changing or incremental data regimes, in some cases they fail completely or suffer from rapid performance degradation on earlier learned tasks, known as catastrophic forgetting. What might be gained by developing neural network models that learn sequentially like humans? First of all, many applications could benefit from continual adaptation to a changing target specification: for example, visual recognition algorithms that need to learn a diverse, growing set of image classes; or household robots that need to incrementally add skills to their repertoire. Continual learning techniques could enable models to acquire specialized solutions without forgetting previous ones, potentially learning over a lifetime, as a human does. In fact, continual learning is generally considered one of the attributes necessary for human-level artificial general intelligence [1]. More fundamentally, continual learning methods could offer enormous advantages for deep neural networks even in stationary settings, by improving learning efficiency as well as by enabling knowledge transfer between related tasks. This article will first motivate a taxonomy of continual learning approaches through describing their connections with biological systems. Just as continual learning in humans cann...

No abstract