Visualizing and Understanding Recurrent Networks
read the original abstract
Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study.
This paper has not been read by Pith yet.
Forward citations
Cited by 8 Pith papers
-
Toy Models of Superposition
Toy models demonstrate that polysemanticity arises when neural networks store more sparse features than neurons via superposition, producing a phase transition tied to polytope geometry and increased adversarial vulne...
-
Letting the neural code speak: Automated characterization of monkey visual neurons through human language
Natural-language descriptions generated and verified through generative models and digital twins capture the selectivity of most neurons in macaque V1 and V4.
-
Letting the neural code speak: Automated characterization of monkey visual neurons through human language
Natural language descriptions generated via a closed-loop pipeline with digital twins capture the selectivity of most neurons in macaque V1 and V4, with synthesized images driving 96% of V4 neurons into the top or bot...
-
Federated Cross-Modal Retrieval with Missing Modalities via Semantic Routing and Adapter Personalization
RCSR is a personalization-friendly federated framework that improves cross-modal retrieval accuracy and stability under missing modalities via semantic routing and adapters.
-
Interpretable and Steerable Sequence Learning via Prototypes
ProSeNet learns a sparse set of prototypes for case-based explanations in deep sequence models, matches state-of-the-art accuracy on several tasks, and supports manual prototype refinement by non-experts.
-
Widening the Representation Bottleneck in Neural Machine Translation with Lexical Shortcuts
Gated lexical shortcut connections added to the transformer yield 0.9 BLEU average gains on five WMT directions while lowering the lexical content stored in hidden states.
-
Investigating Action Encodings in Recurrent Neural Networks in Reinforcement Learning
The authors compare multiple methods for incorporating action information into RNN state updates for RL and report empirical results on illustrative domains.
-
Understanding Memory Modules on Learning Simple Algorithms
NTM and stack-augmented networks both generalize on sequence reversal but only the stack model succeeds on arithmetic expressions by monitoring different input categories and applying distinct memory-update policies.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.