Finite precision RNNs have varying computational power, with LSTMs and ReLU-RNNs being more powerful.
problem Understanding the computational limits of finite precision RNNs for language recognition.
method Comparison of different RNN variants with finite precision and linear computation time.
result LSTMs and ReLU-RNNs are strictly stronger than other RNN variants in terms of computational power.
Study shows challenges in converting RNNs to FSMs due to computational complexity.
problem Understanding the equivalence and distance between RNNs and FSMs.
method Computational proofs for equivalence and distance problems between RNNs and FSMs.
result Undecidability and hardness of approximation problems between RNNs and FSMs.
Paper proposes quantizing RNNs to save space and power.
problem Over-parameterization in RNNs leads to inefficiency.
method Increases bit-width reduction for accuracy preservation.
result RNNs can maintain accuracy with reduced precision.
GCRN models graph-structured sequences with CNN and RNN.
problem Predicting structured sequences of data.
method Combines CNN and RNN on graphs.
result Improves precision and learning speed.
Paper introduces a technique to simplify RNN policies for better understanding and analysis.
problem Difficulty in explaining and analyzing RNN policies due to continuous-valued memory vectors and observation features.
method Quantized Bottleneck Insertion technique to learn finite representations of RNN vectors and features.
result Finite representations of RNN policies can be as small as 3 discrete memory states and 10 observations, improving interpretability.
State-regularized RNNs improve interpretability and performance on long-term memory tasks.
problem RNNs struggle with long-term memory and lack of interpretability.
method Introduce a stochastic state transition mechanism to limit state transitions to a finite set.
result State-regularized RNNs perform better on tasks requiring long-term memory.
This research unifies concepts of fading memory in RNNs.
problem Unclear relationships between fading memory concepts in RNNs.
method Unified language and new proofs for fading memory concepts.
result Clarified relationships between fading memory concepts.
W-RNN improves text classification by extracting serialized text semantics.
problem Semantic constraint in sparse representation classification methods.
method Weighted RNN using word vectors and recurrent neural networks.
result W-RNN outperforms other methods in precision, recall, F1, and loss values.
Linear RNNs exhibit a bias towards shorter memory due to initialization variance.
problem Understanding the performance limitations of RNNs, especially linear ones.
method Kernel regime analysis to show equivalence to 1D-convolutional networks and analyze weightings.
result Linear RNNs with random initialization have a bias towards shorter memory periods.
Paper connects WFA and 2-RNNs, offering a new learning algorithm.
problem Expressiveness and learning of recurrent neural networks.
method Spectral learning algorithm for linear 2-RNNs.
result Provable learning algorithm for linear 2-RNNs.
Study proves deep narrow RNNs can approximate any function, with minimum width independent of data length.
problem Proving universality of deep narrow RNNs with bounded widths.
method Analyzing RNNs as dynamical systems, proving universality for deep narrow structures with specific widths.
result Minimum width for universality of deep narrow RNNs is independent of data length.
New neural stack and Turing Machine architectures prove stability and computational power.
problem Designing stable neural network architectures for Turing Machine simulation.
method Introducing neural stack and Turing Machine architectures, proving stability and computational equivalence.
result Differentiable nnTM with bounded neurons can simulate Turing Machine in real-time and is equivalent to UTM.
A new RNN model based on coupled oscillators mitigates gradient issues.
problem Gradient vanishing and exploding issues in RNNs.
method Time-discretization of a system of second-order ODEs modeling coupled oscillators.
result The model maintains bounded gradients, leading to stable learning of long-term dependencies.
Recurrent neural networks (RNNs), particularly long short-term memory (LSTM), have gained much attention in automatic speech recognition (ASR). Although some successful stories have been reported, training RNNs remains highly challenging, especially with limited training data. Recent research found that a well-trained …
Explains RNN and LSTM fundamentals, derives formulas, and addresses training issues.
problem Lack of detailed formulas and unrolling techniques in LSTM and RNN literature.
method Derives canonical RNN and LSTM formulas from differential equations, proposes unrolling technique, addresses training difficulties.
result Provides a comprehensive understanding of RNN and LSTM, including detailed formulas and unrolling techniques.
Sparse regularization reduces NLP model complexity without sacrificing accuracy.
problem Excessive parameter usage in neural models for NLP leads to high memory and runtime usage.
method Apply group lasso to rational RNNs to learn sparse, parameter-efficient models.
result Sparse rational RNNs can have significantly fewer parameters than non-sparse models without losing performance.
Reservoir computers and RNNs fall short of optimal prediction for stochastic PDFA.
problem Predicting stochastic processes generated by probabilistic deterministic finite-state automata.
method Generalized linear models, Reservoir computers, and Long Short-Term Memory (LSTM) RNNs were tested.
result Each method can fall short of maximal predictive accuracy by up to 50% after training.
Method extracts weighted automata from RNNs using state space regression.
problem Extracting weighted automata from RNNs for better model understanding.
method Regression on RNN state space to prioritize counterexample candidates.
result Quantitative/weighted extension of DFA extraction.
Paper verifies RNNs using automata learning and model checking.
problem Verifying the correctness of RNNs is challenging.
method Learn a deterministic finite automaton from RNN, use model checking for verification.
result Can discover and generalize counterexamples to faulty flows.
Binary and ternary weights simplify RNNs for mobile devices.
problem Complexity and memory intensity of RNNs on mobile devices.
method Learn binary and ternary weights during training.
result Significant memory saving and inference speedup on ASIC platform.
Trains large RNN model on 40GB of text in 4 hours.
problem Training large-scale language models efficiently.
method Mixed precision arithmetic, 32k batch size, 128 NVIDIA GPUs, learning rate schedule.
result Trains character-level LSTM over 40GB of Amazon Reviews in 4 hours.
RNNs learn device models from input/output data.
problem Learning complex device models from limited data.
method Empirical study using RNNs to model six different devices.
result RNNs can generate functional software-only models of hardware devices.
SKOLR uses linear RNNs to approximate Koopman operators for time-series forecasting.
problem Nonlinear dynamical system analysis and time-series forecasting with infinite-dimensional Koopman operators.
method Established a connection between Koopman operator approximation and linear RNNs, integrating learnable spectral decomposition and MLP.
result SKOLR delivers exceptional performance in various forecasting benchmarks and dynamical systems.
PV-RNN model uses predictive coding to predict and recognize data sequences.
problem Addressing variational RNNs' challenges in learning meaningful latent representations and transferring future observations.
method Inspired by predictive coding, PV-RNN dynamically changes latent stochasticity and uses prediction errors for information transfer.
result PV-RNN outperforms standard variational Bayes models in predicting unseen sequences and generalizes well.
Study on memory effects in RNNs learning temporal data.
problem Understanding memory effects in RNNs for temporal data learning.
method Mathematical analysis of continuous-time linear RNNs, focusing on approximation and optimization dynamics.
result Long-term memory requires a large number of neurons and slows down training.
Improves RNN performance under noisy computations.
problem Power and speed limitations in deep learning with noisy analog circuits.
method Deep Noise Injection training to robustify RNN weights/biases.
result Trained RNNs show more consistent performance under noisy inference.
New LTC RNNs can approximate any continuous system with fewer units.
problem Approximating continuous dynamical systems with neural networks.
method Introducing LTC RNNs with variable time-constant synaptic transmission.
result LTC RNNs can approximate any n-dimensional continuous dynamical system. A wearable device-based sleep stage classifier using feature learning and RNNs.
problem Automatic sleep stage classification using wearable devices.
method Multi-level feature learning framework and RNN classifier with BLSTM.
result The algorithm achieves high precision, recall, and F1 scores in both resting and comprehensive groups.
Mathematical study of learning long-term integration in linear RNNs.
problem How do linear recurrent neural networks learn to integrate over long timescales?
method Analytical study of linear RNNs trained to integrate white noise and damped oscillatory filters.
result Learning dynamics are described by low-dimensional effective equations for outlier eigenvalues.
Delayed-RNN approximates stacked and bidirectional RNNs.
problem Improving RNN expressiveness and representational capacity.
method Weight-constrained delayed-RNN, equivalent to stacked-RNNs, with partial acausality.
result Delayed-RNN can approximate stacked and bidirectional RNNs, outperforming them in some tasks.
Paper uses RNNs to design LDPC codes for binary erasure channels.
problem Designing capacity-approaching LDPC codes for binary erasure channels.
method Model Density Evolution using RNNs to determine LDPC code coefficients and structure.
result NDE improves LDPC design performance and complexity compared to differential evolution.
Algorithm extracts deterministic PDFA from probabilistic models with improved performance.
problem Learning deterministic models from probabilistic ones with noise.
method Adapted L* algorithm for probabilistic settings, using conditional probabilities and local tolerance.
result Achieves better performance on WER and NDCG than spectral extraction of WFAs.
Paper derives PAC-Bayesian bounds for LTI systems learning from empirical data.
problem Characterizing predictive power of LTI systems learned from data.
method PAC-Bayesian bounds for LTI stochastic dynamical systems with inputs.
result Finite-sample error bounds for learning algorithms of LTI systems.
Modeling event sequences with RNNs for predictive maintenance.
problem Predicting the intensity function of asynchronous event sequences.
method Use two RNNs: one for background and another for history effects.
result End-to-end training of the model for black-box event intensity prediction.
Transformers solve parity problems efficiently with step-by-step reasoning.
problem Training transformers to solve complex, recursive problems like parity.
method Training a one-layer transformer to solve k-parity, incorporating intermediate parities into the loss function, and using teacher forcing or augmented data. result Transformers can learn parity in one gradient update with intermediate supervision or self-consistency checks.
Finite-precision learning of anh networks is limited by the Monte Carlo rate.
problem Learning anh neural networks under finite precision method Using iterated anh activations to construct localized bump functions result No adaptive randomized algorithm can achieve higher convergence rate than Monte Carlo rate in finite precision
Compact RNNs reduce parameters and improve efficiency.
problem High computational cost of RNNs with large inputs.
method Block-Term Tensor Decomposition (BT-TD) to reduce RNN parameters.
result BT-RNN achieves better accuracy and faster convergence than standard RNNs.
PF-RNNs use particle filtering to model uncertainty in RNNs for better sequential data prediction.
problem Highly variable and noisy sequential data.
method PF-RNNs maintain a latent state distribution as a set of particles, updating with Bayes rule.
result PF-RNNs outperform standard RNNs on various sequence prediction tasks.
In this paper, we explore different ways to extend a recurrent neural network (RNN) to a \textit{deep} RNN. We start by arguing that the concept of depth in an RNN is not as clear as it is in feedforward neural networks. By carefully analyzing and understanding the architecture of an RNN, however, we find three points …
This paper studies a theoretical pruning method for RNNs to reduce computational costs.
problem High computational costs in recurrent neural networks (RNNs).
method Spectral pruning inspired approach for RNNs.
result Generalization error bounds for compressed RNNs are provided.
Lyapunov analysis improves RNN performance prediction.
problem Uncertainty in RNN performance prediction due to hyperparameters and architecture.
method Lyapunov spectral analysis of RNNs and Autoencoder-Lyapunov Embedding Learning (AeLLE).
result AeLLE successfully correlates RNN Lyapunov spectrum with accuracy and predicts performance.
In this work, we propose a novel recurrent neural network (RNN) architecture. The proposed RNN, gated-feedback RNN (GF-RNN), extends the existing approach of stacking multiple recurrent layers by allowing and controlling signals flowing from upper recurrent layers to lower layers using a global gating unit for each pai…
Many machine learning tasks can be expressed as the transformation---or \emph{transduction}---of input sequences into output sequences: speech recognition, machine translation, protein secondary structure prediction and text-to-speech to name but a few. One of the key challenges in sequence transduction is learning to …
Proposes Fusion Recurrent Neural Network for sequence data.
problem Improving sequence learning for practical applications.
method Fusion module and Transport module for sequence data.
result Fusion RNN performs comparably to state-of-the-art RNNs.
MinimalRNN simplifies RNNs for better interpretability and efficiency.
problem Improving interpretability and efficiency of RNNs.
method MinimalRNN uses a simplified structure with minimal updates, leading to efficient learning and testing.
result MinimalRNN learns disentangled RNN states and captures longer range dependencies.
Recurrent neural networks' hidden state can be reconstructed from its past, providing a theoretical framework for stability and tracking.
problem Hidden-state stability in RNNs
method Backward coherence analysis
result Almost-sure convergence, rates under mixing, interpretable limiting representation, finite pathwise stopping times, and theoretical framework for time-uniform confidence sequences.
Interpretable RNN uses sparse recovery for better performance.
problem Interpreting the internal workings of RNNs.
method Sequential Sparse Recovery + SISTA algorithm.
result SISTA-RNN achieves better performance and is more interpretable.
Modified LSTM cells iteratively process input, improving model performance.
problem Improving LSTM network performance for RNNs.
method Iterative scheme where LSTM computations are repeated over constant input and cell state values, updating hidden state multiple times.
result Enhanced model performance, 3 times more parameters than original model.