Neural Information Processing Systems http://nips.cc/

Transformers with linearised attention ("linear Transformers") have demonstrated However, the original FWP formulation is more general than the one of linear Transformers: a slow neural network (NN) continually reprograms the weights of a fast NN with arbitrary architecture. In existing linear Transformers, both NNs are feedforward and consist of a single layer.

Neural Information Processing Systems http://nips.cc/

Transformers with linearised attention ("linear Transformers") have demonstrated However, the original FWP formulation is more general than the one of linear Transformers: a slow neural network (NN) continually reprograms the weights of a fast NN with arbitrary architecture. In existing linear Transformers, both NNs are feedforward and consist of a single layer.

Do autoregressive Transformer language models require explicit positional encodings (PEs)? The answer is "no" as long as they have more than one layer -- they can distinguish sequences with permuted tokens without requiring explicit PEs. This property has been known since early efforts (those contemporary with GPT-2) adopting the Transformer for language modeling. However, this result does not appear to have been well disseminated and was even rediscovered recently. This may be partially due to a sudden growth of the language modeling community after the advent of GPT-2, but perhaps also due to the lack of a clear explanation in prior publications, despite being commonly understood by practitioners in the past. Here we review this long-forgotten explanation why explicit PEs are nonessential for multi-layer autoregressive Transformers (in contrast, one-layer models require PEs to discern order information of their input tokens). We also review the origin of this result, and hope to re-establish it as a common knowledge.

Since the earliest proposals for neural network models of the mind and brain, critics have pointed out key weaknesses in these models compared to human cognitive abilities. Here we review recent work that uses metalearning to overcome several classic challenges by addressing the Problem of Incentive and Practice -- that is, providing machines with both incentives to improve specific skills and opportunities to practice those skills. This explicit optimization contrasts with more conventional approaches that hope the desired behavior will emerge through optimizing related but different objectives. We review applications of this principle to addressing four classic challenges for neural networks: systematic generalization, catastrophic forgetting, few-shot learning and multi-step reasoning. We also discuss the prospects for understanding aspects of human development through this framework, and whether natural environments provide the right incentives and practice for learning how to make challenging generalizations.

An agent trained within a closed system can master any desired capability, as long as the following three conditions hold: (a) it receives sufficiently informative and aligned feedback, (b) its coverage of experience/data is broad enough, and (c) it has sufficient capacity and resource. In this position paper, we justify these conditions, and consider what limitations arise from (a) and (b) in closed systems, when assuming that (c) is not a bottleneck. Considering the special case of agents with matching input and output spaces (namely, language), we argue that such pure recursive self-improvement, dubbed "Socratic learning", can boost performance vastly beyond what is present in its initial data or knowledge, and is only limited by time, as well as gradual misalignment concerns. Furthermore, we propose a constructive framework to implement it, based on the notion of language games.