Spike No More: Stabilizing the Pre-training of Large Language Models

The loss spike often occurs during pre-training of a large language model. The spikes degrade the performance of a large language model, and sometimes ruin the pre-training. Since the pre-training needs a vast computational budget, we should avoid such spikes. To investigate a cause of loss spikes, we focus on gradients of internal layers in this study. Through theoretical analyses, we introduce two causes of the exploding gradients, and provide requirements to prevent the explosion. In addition, we introduce the combination of the initialization method and a simple modification to embeddings as a method to satisfy the requirements. We conduct various experiments to verify our theoretical analyses empirically. Experimental results indicate that the combination is effective in preventing spikes during pre-training.

On Layer Normalizations and Residual Connections in Transformers

In the perspective of a layer normalization (LN) position, the architecture of Transformers can be categorized into two types: Post-LN and Pre-LN. Recent Transformers prefer to select Pre-LN because the training in Post-LN with deep Transformers, e.g., ten or more layers, often becomes unstable, resulting in useless models. However, in contrast, Post-LN has also consistently achieved better performance than Pre-LN in relatively shallow Transformers, e.g., six or fewer layers. This study first investigates the reason for these discrepant observations empirically and theoretically and discovers 1, the LN in Post-LN is the source of the vanishing gradient problem that mainly leads the unstable training whereas Pre-LN prevents it, and 2, Post-LN tends to preserve larger gradient norms in higher layers during the back-propagation that may lead an effective training. Exploiting the new findings, we propose a method that can equip both higher stability and effective training by a simple modification from Post-LN. We conduct experiments on a wide range of text generation tasks and demonstrate that our method outperforms Pre-LN, and stable training regardless of the shallow or deep layer settings.

Decomposing NeRF for Editing via Feature Field Distillation

Emerging neural radiance fields (NeRF) are a promising scene representation for computer graphics, enabling high-quality 3D reconstruction and novel view synthesis from image observations. However, editing a scene represented by a NeRF is challenging, as the underlying connectionist representations such as MLPs or voxel grids are not object-centric or compositional. In particular, it has been difficult to selectively edit specific regions or objects. In this work, we tackle the problem of semantic scene decomposition of NeRFs to enable query-based local editing of the represented 3D scenes. We propose to distill the knowledge of off-the-shelf, self-supervised 2D image feature extractors such as CLIP-LSeg or DINO into a 3D feature field optimized in parallel to the radiance field. Given a user-specified query of various modalities such as text, an image patch, or a point-and-click selection, 3D feature fields semantically decompose 3D space without the need for re-training and enable us to semantically select and edit regions in the radiance field. Our experiments validate that the distilled feature fields (DFFs) can transfer recent progress in 2D vision and language foundation models to 3D scene representations, enabling convincing 3D segmentation and selective editing of emerging neural graphics representations.

Diverse Lottery Tickets Boost Ensemble from a Single Pretrained Model

Ensembling is a popular method used to improve performance as a last resort. However, ensembling multiple models finetuned from a single pretrained model has been not very effective; this could be due to the lack of diversity among ensemble members. This paper proposes Multi-Ticket Ensemble, which finetunes different subnetworks of a single pretrained model and ensembles them. We empirically demonstrated that winning-ticket subnetworks produced more diverse predictions than dense networks, and their ensemble outperformed the standard ensemble on some tasks.

Instance-Based Neural Dependency Parsing

Interpretable rationales for model predictions are crucial in practical applications. We develop neural models that possess an interpretable inference process for dependency parsing. Our models adopt instance-based inference, where dependency edges are extracted and labeled by comparing them to edges in a training set. The training edges are explicitly used for the predictions; thus, it is easy to grasp the contribution of each edge to the predictions. Our experiments show that our instance-based models achieve competitive accuracy with standard neural models and have the reasonable plausibility of instance-based explanations.

SHAPE: Shifted Absolute Position Embedding for Transformers

Position representation is crucial for building position-aware representations in Transformers. Existing position representations suffer from a lack of generalization to test data with unseen lengths or high computational cost. We investigate shifted absolute position embedding (SHAPE) to address both issues. The basic idea of SHAPE is to achieve shift invariance, which is a key property of recent successful position representations, by randomly shifting absolute positions during training. We demonstrate that SHAPE is empirically comparable to its counterpart while being simpler and faster.

Efficient Estimation of Influence of a Training Instance

Understanding the influence of a training instance on a neural network model leads to improving interpretability. However, it is difficult and inefficient to evaluate the influence, which shows how a model's prediction would be changed if a training instance were not used. In this paper, we propose an efficient method for estimating the influence. Our method is inspired by dropout, which zero-masks a sub-network and prevents the sub-network from learning each training instance. By switching between dropout masks, we can use sub-networks that learned or did not learn each training instance and estimate its influence. Through experiments with BERT and VGGNet on classification datasets, we demonstrate that the proposed method can capture training influences, enhance the interpretability of error predictions, and cleanse the training dataset for improving generalization.

All Word Embeddings from One Embedding

In neural network-based models for natural language processing (NLP), the largest part of the parameters often consists of word embeddings. Conventional models prepare a large embedding matrix whose size depends on the vocabulary size. Therefore, storing these models in memory and disk storage is costly. In this study, to reduce the total number of parameters, the embeddings for all words are represented by transforming a shared embedding. The proposed method, ALONE (all word embeddings from one), constructs the embedding of a word by modifying the shared embedding with a filter vector, which is word-specific but non-trainable. Then, we input the constructed embedding into a feed-forward neural network to increase its expressiveness. Naively, the filter vectors occupy the same memory size as the conventional embedding matrix, which depends on the vocabulary size. To solve this issue, we also introduce a memory-efficient filter construction approach. We indicate our ALONE can be used as word representation sufficiently through an experiment on the reconstruction of pre-trained word embeddings. In addition, we also conduct experiments on NLP application tasks: machine translation and summarization. We combined ALONE with the current state-of-the-art encoder-decoder model, the Transformer, and achieved comparable scores on WMT 2014 English-to-German translation and DUC 2004 very short summarization with less parameters.

Instance-Based Learning of Span Representations: A Case Study through Named Entity Recognition

Interpretable rationales for model predictions play a critical role in practical applications. In this study, we develop models possessing interpretable inference process for structured prediction. Specifically, we present a method of instance-based learning that learns similarities between spans. At inference time, each span is assigned a class label based on its similar spans in the training set, where it is easy to understand how much each training instance contributes to the predictions. Through empirical analysis on named entity recognition, we demonstrate that our method enables to build models that have high interpretability without sacrificing performance.

Data Interpolating Prediction: Alternative Interpretation of Mixup

Data augmentation by mixing samples, such as Mixup, has widely been used typically for classification tasks. However, this strategy is not always effective due to the gap between augmented samples for training and original samples for testing. This gap may prevent a classifier from learning the optimal decision boundary and increase the generalization error. To overcome this problem, we propose an alternative framework called Data Interpolating Prediction (DIP). Unlike common data augmentations, we encapsulate the sample-mixing process in the hypothesis class of a classifier so that train and test samples are treated equally. We derive the generalization bound and show that DIP helps to reduce the original Rademacher complexity. Also, we empirically demonstrate that DIP can outperform existing Mixup.