| Challenge: | Existing approaches to model long-term dependencies are limited to long texts with thousands of words. |
| Approach: | They propose a look-ahead memory that augments the recurrence memory by attending to the right-side tokens and interpolating with the old memory states to maintain long-term information in the history. |
| Outcome: | Experiments on widely used language modeling benchmarks show that LaMemo outperforms baseline models with recurrence memory. |
Similar Papers
Interpreting Context Look-ups in Transformers: Investigating Attention-MLP Interactions (2024.emnlp-main)
Copied to clipboard
| Challenge: | Using a method to identify next-token neurons, we find that some attention heads recognize contexts relevant to predicting a token and activate a downstream token-predicting neuron accordingly. |
| Approach: | They propose a method to identify next-token neurons and determine the upstream attention heads responsible for their activity in LLMs. |
| Outcome: | The proposed method identifies next-token neurons, finds prompts that highly activate them, and determines the upstream attention heads responsible. |
HMT: Hierarchical Memory Transformer for Efficient Long Context Language Processing (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing models that memorize past tokens have “flat” memory architectures that restrict the context window. |
| Approach: | They propose a framework that imitates human memorization behavior by preserving tokens from early input segments, passing memory embeddings along the sequence, and recalling relevant information from history. |
| Outcome: | The proposed framework outperforms existing models in language modeling and question-answering tasks and achieves comparable or superior generation quality to long-context models with 2 57 fewer parameters and 2.5 116 less inference memory. |
Adaptive Attention Span in Transformers (P19-1)
Copied to clipboard
| Challenge: | We extend the maximum context size of a neural network called Transformer to 8k characters. |
| Approach: | They propose a self-attention mechanism that can learn its optimal attention span . this allows for models with longer context and the capability to catch longer dependencies. |
| Outcome: | The proposed model achieves state-of-the-art performance on text8 and enwiki8 using 8k characters with no loss of performance, and maintains control over memory footprint and computational time. |
Do Long-Range Language Models Actually Use Long-Range Context? (2021.emnlp-main)
Copied to clipboard
| Challenge: | Language models are generally trained on short, truncated input sequences, which limits their ability to use discourse-level information present in long-range context to improve their predictions. |
| Approach: | They analyze two long-range Transformer language models that accept 8K token inputs . they find that providing long-term context only improves their predictions on a small set of tokens - not sentence-level ones . |
| Outcome: | The proposed model improves on PG-19 with only 2K tokens and does not help at all for sentence-level prediction tasks. |
Flashback: Memory Mechanism for Enhancing Memory Efficiency and Speed in Deep Sequential Models (2025.coling-main)
Copied to clipboard
| Challenge: | Existing deep sequential processing models have problems with memory degradation and inaccurate gradient backpropagation. |
| Approach: | They propose a Flashback property that preserves memory as an identity mapping until it is overwritten by a hidden state at a different time step. |
| Outcome: | The proposed model can be implemented in Transformers and Mamba, and it performs well. |
FragRel: Exploiting Fragment-level Relations in the External Memory of Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to process contexts with unlimited length are limited to finite expansion length or prone to performance degradation when dealing with very long contexts. |
| Approach: | They propose to exploit fragment-level relations in external memory to hierarchically process the long text. |
| Outcome: | The proposed model improves story understanding, repository-level code generation, and long-term chatting. |
MemoryPrompt: A Light Wrapper to Improve Context Tracking in Pre-trained Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Transformer-based language models (LMs) track contextual information through large, hard-coded input windows. |
| Approach: | They propose a leaner approach where a pre-trained LM is augmented with a small auxiliary recurrent network that passes information to the LM by prefixing its regular input with . vectors. |
| Outcome: | The proposed method outperforms larger LMs with full input history on a long-distance dialogue dataset and does not suffer catastrophic forgetting when adapted to new tasks. |
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. |
Insights into LLM Long-Context Failures: When Transformers Know but Don’t Tell (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) exhibit positional bias, struggling to utilize information from the middle or end of long contexts. |
| Approach: | They propose to examine LLMs' long-context generalizations by probing their hidden representations. |
| Outcome: | The proposed models excel at processing extended contexts while preserving their positional bias. |
ABC: Attention with Bounded-memory Control (2022.acl-long)
Copied to clipboard
Hao Peng, Jungo Kasai, Nikolaos Pappas, Dani Yogatama, Zhaofeng Wu, Lingpeng Kong, Roy Schwartz, Noah A. Smith
| Challenge: | Existing approaches to attention with bounded-memory control (ABC) have a quadratic complexity in sequence lengths, making it prohibitive for long sequences. |
| Approach: | They propose a new abstraction that bounds memory size to improve efficiency . they propose bounded-memory control, which connects several efficient attention variants . |
| Outcome: | The proposed approach outperforms existing approaches on language modeling, machine translation, and masked language model finetuning. |