Papers with WikiText-103
Relational Memory-Augmented Language Models (2022.tacl-1)
Copied to clipboard
| Challenge: | Existing language models rely on word correlation and are difficult to interpret . existing models often lack explicit representations for such information . |
| Approach: | They propose a memory-augmented approach to condition autoregressive language models on knowledge graphs. |
| Outcome: | The proposed model improves perplexity and bits per character in an autoregressive language model . it is complementary to token-based memory and enables causal interventions . |
Approximating Two-Layer Feedforward Networks for Efficient Transformers (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent work uses sparse Mixtures of Experts (MoEs) to build resource-efficient large language models. |
| Approach: | They propose a general framework that unifies various methods to build two-layer NNs . they propose methods to improve both MoEs and PKMs based on their results . |
| Outcome: | The proposed framework improves both MoEs and product-key memories (PKMs) it shows that MoE's are competitive with dense Transformer-XL on two different scales while being much more resource efficient. |
Better Language Model with Hypernym Class Prediction (2022.acl-long)
Copied to clipboard
| Challenge: | Class-based language models (LMs) have been devised to address context sparsity in n-gram LMs for decades. |
| Approach: | They propose to use class-based prediction to improve generalization for rare words by annealing from predicting the class to token prediction during training. |
| Outcome: | The proposed model improves perplexity without sacrificing performance on rare words. |
Pushdown Layers: Encoding Recursive Structure in Transformer Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Pushdown Layers model recursive state via stack tape that tracks estimated depths of tokens in incremental parsing . pushdown layers are drop-in replacement for standard self-attention . recursion is a key component of many aspects of intelligent behavior, authors say . |
| Approach: | They propose a self-attention layer that models recursive state via a stack tape . Pushdown Layers is a drop-in replacement for standard self- attention . |
| Outcome: | The proposed self-attention layer improves on parse tasks with a recursive-state model . it can model recursion using a stack tape that tracks estimated depths of tokens . |
ERNIE-Doc: A Retrospective Long-Document Modeling Transformer (2021.acl-long)
Copied to clipboard
| Challenge: | Existing models for document-level language pretraining are not suitable for long documents due to their quadratically increasing memory and time consumption. |
| Approach: | They propose a document-level language pretraining model based on Recurrence Transformers. |
| Outcome: | The proposed model outperforms existing models on language understanding tasks. |
Transformer-XL: Attentive Language Models beyond a Fixed-Length Context (P19-1)
Copied to clipboard
| Challenge: | Term memory networks (RNNs) are difficult to optimize due to gradient vanishing and explosion. |
| Approach: | They propose a neural architecture Transformer-XL that enables learning dependency beyond a fixed length without disrupting temporal coherence. |
| Outcome: | The proposed method improves state-of-the-art performance on short and long sequences and generates coherent, novel text articles with thousands of tokens. |
TRAMS: Training-free Memory Selection for Long-range Language Modeling (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods like Transformer-XL are plagued by ineffective memory selections due to the high number of tokens involved in attention calculation. |
| Approach: | They propose a plug-and-play strategy that selects tokens participating in attention calculation based on one simple metric and ignores the other ones. |
| Outcome: | The proposed strategy keeps tokens with high attention scores and ignores the other ones on word-level and character-level benchmarks without additional training or adding additional parameters. |
Reconsidering the Past: Optimizing Hidden States in Language Models (2021.findings-emnlp)
Copied to clipboard
| Challenge: | In this paper, we present Hidden-State Optimization (HSO) for language models at inference time. |
| Approach: | They propose a method that uses the log-probability gradient to update hidden states rather than the model parameters to improve the performance of transformer language models. |
| Outcome: | The proposed method improves performance of transformer language models at inference time. |
Sneaking Syntax into Transformer Language Models with Tree Regularization (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods for incorporating syntactic inductive biases into transformers are limited . we introduce auxiliary loss function that converts bracketing decisions into differentiable orthogonality constraints on vector hidden states. |
| Approach: | They propose to introduce syntactic inductive biases into transformer circuits through a structured regularizer. |
| Outcome: | The proposed approach could unlock more robust and data-efficient learning in transformer language models . it integrates seamlessly with the standard LM objective, requiring no architectural changes. |
Shortformer: Better Language Modeling using Shorter Inputs (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods require computationally expensive relative position embeddings. |
| Approach: | They propose two methods that decrease input length to improve perplexity and perplexability. |
| Outcome: | The proposed methods speed up training by a factor of 1.65 and reduce memory usage. |
Efficient Unstructured Pruning of Mamba State-Space Models for Resource-Constrained Environments (2025.emnlp-main)
Copied to clipboard
| Challenge: | State-space models struggle with quadratic computational complexity, limiting their use in long-context tasks and resource-constrained input data. |
| Approach: | They propose a pruning framework specifically tailored for Mamba that reduces parameter counts by 70% with only a 3–9% drop in performance. |
| Outcome: | The proposed pruning framework achieves up to 70% parameter reduction with only a 3–9% drop in performance. |
Fine-Tuning Pre-trained Transformers into Decaying Fast Weights (2022.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive Transformers incur O(T) complexity during per-token generation due to the self-attention mechanism. |
| Approach: | They propose a kernel-based method to approximate causal self-attention by replacing it with recurrent formulations with various update rules and feature maps to achieve O(1) time and memory complexity. |
| Outcome: | The proposed method outperforms prior methods and retains 99% of attention’s performance on WikiText-103 against more complex attention substitutes. |
PermuteFormer: Efficient Relative Position Encoding for Long Sequences (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing Transformers that scale to long sequences are not compatible with relative position encoding. |
| Approach: | They propose a Performer-based model with relative position encoding that scales linearly on long sequences. |
| Outcome: | The proposed model outperforms performer on long sequences with no computational overhead and outperformed vanilla Transformer on most of the tasks. |
BSFA: Leveraging the Subspace Dichotomy to Accelerate Neural Network Training (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies highlight a fundamental dichotomy in deep learning optimization: parameter updates along the top eigendirections of the loss Hessian (Dom-space) capture most of the update magnitude, while updates in the orthogonal component (Bulk-space), have smaller magnitudes but drive most learning progress. |
| Approach: | They propose a plug-and-play framework that scales update components projected onto distinct subspaces and a block-wise strategy that applies this estimation on a per-parameter-block basis. |
| Outcome: | The proposed framework accelerates training by differentially scaling update components projected onto distinct subspaces, while enhancing stability by moderating updates in dominant subspace and boosting convergence speed by amplifying updates in bulk-space. |
Pruning before Fine-tuning: A Retraining-free Compression Framework for Pre-trained Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Structured pruning is an effective technique for compressing pre-trained language models (PLMs), but it requires retraining, leading to additional computational overhead. |
| Approach: | They propose a task-specific pruning framework that prunes redundant modules of pre-trained language models before fine-tuning them. |
| Outcome: | The proposed pruning framework achieves higher performance on GLUE, SQUAD, WikiText-2, Wik-103, and PTB datasets while reducing the time required for fine-tuning. |