| Challenge: | Existing methods for word embedding reconstruction use only local information of subwords and pre-trained word embeds. |
| Approach: | They propose a global loss function that uses words other than the target word to improve word embedding reconstruction by a factor of 200. |
| Outcome: | The proposed method reduces the model size of pre-trained word embeddings by a factor of 200 while preserving its quality. |
Similar Papers
Subword-based Compact Reconstruction of Word Embeddings (N19-1)
Copied to clipboard
| Challenge: | Existing word-based word embeddings are based on subword information and memory-shared embeddables. |
| Approach: | They propose a method for reconstructing pre-trained word embeddings using subword information using memory-shared embedds and a variant of the key-value-query self-attention mechanism. |
| Outcome: | The proposed method can imitate well-trained word embeddings in a small fixed space while preventing quality degradation across several linguistic benchmark datasets. |
Smaller Text Classifiers with Discriminative Cluster Embeddings (N18-2)
Copied to clipboard
| Challenge: | Word embeddings dominate overall model sizes in neural methods for natural language processing, especially when large vocabularies and high dimensions are used. |
| Approach: | They propose a Gumbel-Softmax distribution to maximize over the latent clustering while minimizing the task loss. |
| Outcome: | The proposed method minimizes the task loss while maximizing over the latent clustering while remaining parameter-efficient. |
Generalizing Word Embeddings using Bag of Subwords (D18-1)
Copied to clipboard
| Challenge: | Existing word embeddings techniques have a fixed vocabulary, i.e., they can only provide vectors over a finite set of common words that appear frequently in a given corpus. |
| Approach: | They propose a subword-level word vector generation model that views words as bags of character n-grams and provides good vectors for rare or unseen words. |
| Outcome: | The proposed model performs state-of-the-art in English word similarity task and in joint prediction of part-of speech tag and morphosyntactic attributes in 23 languages. |
Reusing Weights in Subword-Aware Neural Language Models (N18-1)
Copied to clipboard
| Challenge: | a statistical language model assigns a probability to a sequence of words . data sparsity is a major problem in building traditional n-gram language models . |
| Approach: | They propose several ways to reuse subword embeddings and other weights in subword-aware neural language models. |
| Outcome: | The proposed techniques do not benefit a competitive character-aware model . but they show significant reductions in model sizes and performance. |
Better Word Embeddings by Disentangling Contextual n-Gram Information (N19-1)
Copied to clipboard
| Challenge: | Pre-trained word vectors are ubiquitous in Natural Language Processing applications. |
| Approach: | They show that word embeddings with bigram and trigram embedds improve unigram embeds . they claim this removes contextual information from unigrammes, resulting in better unigraph embedders . |
| Outcome: | The proposed model outperforms competing models on a wide variety of tasks. |
Autoencoding Improves Pre-trained Word Embeddings (2020.coling-main)
Copied to clipboard
| Challenge: | Existing work has shown that word embeddings are distributed in a narrow cone and that centering and projection can improve the accuracy of pre-trained word embeds without requiring additional training data. |
| Approach: | They propose to remove the top principal components from pre-trained word embeddings and center and project them onto principal component vectors to reinstate isotropy in the embeddable space. |
| Outcome: | The proposed method is equivalent to applying a linear autoencoder to minimize the squared L2 reconstruction error. |
EmbedTextNet: Dimension Reduction with Weighted Reconstruction and Correlation Losses for Efficient Text Embedding (2023.findings-acl)
Copied to clipboard
| Challenge: | EmbedTextNet is a light add-on network that can be appended to an arbitrary language model to generate a compact embedding without requiring any changes in its architecture or training procedure. |
| Approach: | They propose an add-on network that can be appended to an arbitrary language model to generate a compact embedding without requiring any changes in its architecture or training procedure. |
| Outcome: | The proposed network can be appended to an arbitrary language model to generate a compact embedding without any changes in its architecture or training procedure. |
Text Embeddings Reveal (Almost) As Much As Text (2023.emnlp-main)
Copied to clipboard
| Challenge: | a vector database of dense text embeddings stores only the text data, not the original text . a multi-step method that iteratively corrects and re-embeds text can recover 92% of 32-token text inputs exactly. |
| Approach: | They propose a method that iteratively corrects and re-embeds text to recover 92% of 32-token text inputs exactly. |
| Outcome: | The proposed method recovers 92% of 32-token text inputs exactly. |
Advances in Pre-Training Distributed Word Representations (L18-1)
Copied to clipboard
| Challenge: | Pre-trained word representations are a building block of many Natural Language Processing and Machine Learning applications. |
| Approach: | They propose to combine known tricks and a set of publicly available pre-trained word vector representations to train high-quality representations. |
| Outcome: | The proposed models outperform the current state of the art on a number of tasks while maintaining a high training speed to scale to massive amount of data. |
How to represent a word and predict it, too: Improving tied architectures for language modelling (D18-1)
Copied to clipboard
| Challenge: | Recent state-of-the-art models use word embeddings as input and output mappings instead of tied models. |
| Approach: | They propose to decouple hidden state from word embedding prediction . they extend their proposed modification to word2vec models . |
| Outcome: | The proposed architectures achieve comparable or better results compared to previous models without tying . the proposed architecture reduces parameters, enabling more compact models and faster learning. |