Natural Language Understanding

A Transformer is a model architecture that eschews recurrence and instead relies entirely on an attention mechanism to draw global dependencies between input and output. Before Transformers, the dominant sequence transduction models were based on complex recurrent or convolutional neural networks that include an encoder and a decoder. The Transformer also employs an encoder and decoder, but removing recurrence in favor of attention mechanisms allows for significantly more parallelization than methods like RNNs and CNNs.

Absolute Position Encodings are a type of position embeddings for [Transformer-based models] where positional encodings are added to the input embeddings at the bottoms of the encoder and decoder stacks. The positional encodings have the same dimension $d\_{model}$ as the embeddings, so that the two can be summed. In the original implementation, sine and cosine functions of different frequencies are used: $$ \text{PE}\left(pos, 2i\right) = \sin\left(pos/10000^{2i/d\_{model}}\right) $$ $$ \text{PE}\left(pos, 2i+1\right) = \cos\left(pos/10000^{2i/d\_{model}}\right) $$ where $pos$ is the position and $i$ is the dimension. That is, each dimension of the positional encoding corresponds to a sinusoid. The wavelengths form a geometric progression from $2\pi$ to $10000 \dot 2\pi$. This function was chosen because the authors hypothesized it would allow the model to easily learn to attend by relative positions, since for any fixed offset $k$, $\text{PE}\_{pos+k}$ can be represented as a linear function of $\text{PE}\_{pos}$. Image Source: D2L.ai

BERT, or Bidirectional Encoder Representations from Transformers, improves upon standard Transformers by removing the unidirectionality constraint by using a masked language model (MLM) pre-training objective. The masked language model randomly masks some of the tokens from the input, and the objective is to predict the original vocabulary id of the masked word based only on its context. Unlike left-to-right language model pre-training, the MLM objective enables the representation to fuse the left and the right context, which allows us to pre-train a deep bidirectional Transformer. In addition to the masked language model, BERT uses a next sentence prediction task that jointly pre-trains text-pair representations. There are two steps in BERT: pre-training and fine-tuning. During pre-training, the model is trained on unlabeled data over different pre-training tasks. For fine-tuning, the BERT model is first initialized with the pre-trained parameters, and all of the parameters are fine-tuned using labeled data from the downstream tasks. Each downstream task has separate fine-tuned models, even though they are initialized with the same pre-trained parameters.

Contrastive Language-Image Pre-training (CLIP), consisting of a simplified version of ConVIRT trained from scratch, is an efficient method of image representation learning from natural language supervision. , CLIP jointly trains an image encoder and a text encoder to predict the correct pairings of a batch of (image, text) training examples. At test time the learned text encoder synthesizes a zero-shot linear classifier by embedding the names or descriptions of the target dataset’s classes. For pre-training, CLIP is trained to predict which of the $N X N$ possible (image, text) pairings across a batch actually occurred. CLIP learns a multi-modal embedding space by jointly training an image encoder and text encoder to maximize the cosine similarity of the image and text embeddings of the $N$ real pairs in the batch while minimizing the cosine similarity of the embeddings of the $N^2 - N$ incorrect pairings. A symmetric cross entropy loss is optimized over these similarity scores. Image credit: Learning Transferable Visual Models From Natural Language Supervision

DeBERTa is a Transformer-based neural language model that aims to improve the BERT and RoBERTa models with two techniques: a disentangled attention mechanism and an enhanced mask decoder. The disentangled attention mechanism is where each word is represented unchanged using two vectors that encode its content and position, respectively, and the attention weights among words are computed using disentangle matrices on their contents and relative positions. The enhanced mask decoder is used to replace the output softmax layer to predict the masked tokens for model pre-training. In addition, a new virtual adversarial training method is used for fine-tuning to improve model’s generalization on downstream tasks.

XLM is a Transformer based architecture that is pre-trained using one of three language modelling objectives: 1. Causal Language Modeling - models the probability of a word given the previous words in a sentence. 2. Masked Language Modeling - the masked language modeling objective of BERT. 3. Translation Language Modeling - a (new) translation language modeling objective for improving cross-lingual pre-training. The authors find that both the CLM and MLM approaches provide strong cross-lingual features that can be used for pretraining models.

Mixed Attention Block is an attention module used in the ConvBERT architecture. It is a mixture of self-attention and span-based dynamic convolution (highlighted in pink). They share the same Query but use different Key to generate the attention map and convolution kernel respectively. The number of attention heads is reducing by directly projecting the input to a smaller embedding space to form a bottleneck structure for self-attention and span-based dynamic convolution. Dimensions of the input and output of some blocks are labeled on the left top corner to illustrate the overall framework, where $d$ is the embedding size of the input and $\gamma$ is the reduction ratio.

Span-Based Dynamic Convolution is a type of convolution used in the ConvBERT architecture to capture local dependencies between tokens. Kernels are generated by taking in a local span of current token, which better utilizes local dependency and discriminates different meanings of the same token (e.g., if “a” is in front of “can” in the input sentence, “can” is apparently a noun not a verb). Specifically, with classic convolution, we would have fixed parameters shared for all input tokens. Dynamic convolution is therefore preferable because it has higher flexibility in capturing local dependencies of different tokens. Dynamic convolution uses a kernel generator to produce different kernels for different input tokens. However, such dynamic convolution cannot differentiate the same tokens within different context and generate the same kernels (e.g., the three “can” in Figure (b)). Therefore the span-based dynamic convolution is developed to produce more adaptive convolution kernels by receiving an input span instead of only a single token, which enables discrimination of generated kernels for the same tokens within different context. For example, as shown in Figure (c), span-based dynamic convolution produces different kernels for different “can” tokens.

ConvBERT is a modification on the BERT architecture which uses a span-based dynamic convolution to replace self-attention heads to directly model local dependencies. Specifically a new mixed attention module replaces the self-attention modules in BERT, which leverages the advantages of convolution to better capture local dependency. Additionally, a new span-based dynamic convolution operation is used to utilize multiple input tokens to dynamically generate the convolution kernel. Lastly, ConvBERT also incorporates some new model designs including the bottleneck attention and grouped linear operator for the feed-forward module (reducing the number of parameters).

CuBERT, or Code Understanding BERT, is a BERT based model for code understanding. In order to achieve this, the authors curate a massive corpus of Python programs collected from GitHub. GitHub projects are known to contain a large amount of duplicate code. To avoid biasing the model to such duplicated code, authors perform deduplication using the method of Allamanis (2018). The resulting corpus has 7.4 million files with a total of 9.3 billion tokens (16 million unique).

I-BERT is a quantized version of BERT that quantizes the entire inference with integer-only arithmetic. Based on lightweight integer only approximation methods for nonlinear operations, e.g., GELU, Softmax, and Layer Normalization, it performs an end-to-end integer-only BERT inference without any floating point calculation. In particular, GELU and Softmax are approximated with lightweight second-order polynomials, which can be evaluated with integer-only arithmetic. For LayerNorm, integer-only computation is performed by leveraging a known algorithm for integer calculation of square root.

Probabilistically Masked Language Model, or PMLM, is a type of language model that utilizes a probabilistic masking scheme, aiming to bridge the gap between masked and autoregressive language models. The basic idea behind the connection of two categories of models is similar to MADE by Germain et al (2015). PMLM is a masked language model with a probabilistic masking scheme, which defines the way sequences are masked by following a probabilistic distribution. The authors employ a simple uniform distribution of the masking ratio and name the model as u-PMLM.

CPT, or Chinese Pre-trained Unbalanced Transformer, is a pre-trained unbalanced Transformer for Chinese natural language understanding (NLU) and natural language generation (NLG) tasks. CPT consists of three parts: a shared encoder, an understanding decoder, and a generation decoder. Two specific decoders with a shared encoder are pre-trained with masked language modeling (MLM) and denoising auto-encoding (DAE) tasks, respectively. With the partially shared architecture and multi-task pre-training, CPT can (1) learn specific knowledge of both NLU or NLG tasks with two decoders and (2) be fine-tuned flexibly that fully exploits the potential of the model. Two specific decoders with a shared encoder are pre-trained with masked language modeling (MLM) and denoising auto-encoding (DAE) tasks, respectively. With the partially shared architecture and multi-task pre-training, CPT can (1) learn specific knowledge of both NLU or NLG tasks with two decoders and (2) be fine-tuned flexibly that fully exploits the potential of the model.

PanGu-$α$ is an autoregressive language model (ALM) with up to 200 billion parameters pretrained on a large corpus of text, mostly in Chinese language. The architecture of PanGu-$α$ is based on Transformer, which has been extensively used as the backbone of a variety of pretrained language models such as BERT and GPT. Different from them, there's an additional query layer developed on top of Transformer layers which aims to explicitly induce the expected output.

PatchMerger is a module for Vision Transformers that decreases the number of tokens/patches passed onto each individual transformer encoder block whilst maintaining performance and reducing compute. PatchMerger takes linearly transforms an input of shape N patches × D dimensions through a learnable weight matrix of shape M output patches × D. This generates M scores, in which a Softmax function is applied for each score. The resulting output has a shape of M × N, which is multiplied to the original input to get an output of shape M × D. Mathematically, $$Y = \text{softmax}({W^T}{X^T})X$$ Image and formula from: Renggli, C., Pinto, A. S., Houlsby, N., Mustafa, B., Puigcerver, J., & Riquelme, C. (2022). Learning to Merge Tokens in Vision Transformers. arXiv preprint arXiv:2202.12015.

SMITH, or Siamese Multi-depth Transformer-based Hierarchical Encoder, is a Transformer-based model for document representation learning and matching. It contains several design choices to adapt self-attention models for long text inputs. For the model pre-training, a masked sentence block language modeling task is used in addition to the original masked word language model task used in BERT, to capture sentence block relations within a document. Given a sequence of sentence block representation, the document level Transformers learn the contextual representation for each sentence block and the final document representation.

BARThez is a self-supervised transfer learning model for the French language based on BART. Compared to existing BERT-based French language models such as CamemBERT and FlauBERT, BARThez is well-suited for generative tasks, since not only its encoder but also its decoder is pretrained.