Neural scaling laws define a predictable relationship between a model's parameter count and its performance after training in the form of a power law. However, most research to date has not explicitly investigated whether scaling laws can be used to accelerate model development. In this work, we perform such an empirical investigation across a wide range of language understanding tasks, starting from models with as few as 10K parameters, and evaluate downstream performance across 9 language understanding tasks. We find that scaling laws emerge at finetuning time in some NLP tasks, and that they can also be exploited for debugging convergence when training large models. Moreover, for tasks where scaling laws exist, they can be used to predict the performance of larger models, which enables effective model selection. However, revealing scaling laws requires careful hyperparameter tuning and multiple runs for the purpose of uncertainty estimation, which incurs additional overhead, partially offsetting the computational benefits.
相關內容
We introduce Action-GPT, a plug and play framework for incorporating Large Language Models (LLMs) into text-based action generation models. Action phrases in current motion capture datasets contain minimal and to-the-point information. By carefully crafting prompts for LLMs, we generate richer and fine-grained descriptions of the action. We show that utilizing these detailed descriptions instead of the original action phrases leads to better alignment of text and motion spaces. Our experiments show qualitative and quantitative improvement in the quality of synthesized motions produced by recent text-to-motion models. Code, pretrained models and sample videos will be made available at //actiongpt.github.io
Fine-tuning pre-trained models has been ubiquitously proven to be effective in a wide range of NLP tasks. However, fine-tuning the whole model is parameter inefficient as it always yields an entirely new model for each task. Currently, many research works propose to only fine-tune a small portion of the parameters while keeping most of the parameters shared across different tasks. These methods achieve surprisingly good performance and are shown to be more stable than their corresponding fully fine-tuned counterparts. However, such kind of methods is still not well understood. Some natural questions arise: How does the parameter sparsity lead to promising performance? Why is the model more stable than the fully fine-tuned models? How to choose the tunable parameters? In this paper, we first categorize the existing methods into random approaches, rule-based approaches, and projection-based approaches based on how they choose which parameters to tune. Then, we show that all of the methods are actually sparse fine-tuned models and conduct a novel theoretical analysis of them. We indicate that the sparsity is actually imposing a regularization on the original model by controlling the upper bound of the stability. Such stability leads to better generalization capability which has been empirically observed in a lot of recent research works. Despite the effectiveness of sparsity grounded by our theory, it still remains an open problem of how to choose the tunable parameters. To better choose the tunable parameters, we propose a novel Second-order Approximation Method (SAM) which approximates the original problem with an analytically solvable optimization function. The tunable parameters are determined by directly optimizing the approximation function. The experimental results show that our proposed SAM model outperforms many strong baseline models and it also verifies our theoretical analysis.
The ability to extrapolate, i.e., to make predictions on sequences that are longer than those presented as training examples, is a challenging problem for current deep learning models. Recent work shows that this limitation persists in state-of-the-art Transformer-based models. Most solutions to this problem use specific architectures or training methods that do not generalize to other tasks. We demonstrate that large language models can succeed in extrapolation without modifying their architecture or training procedure. Our experimental results show that generating step-by-step rationales and introducing marker tokens are both required for effective extrapolation. First, we induce a language model to produce step-by-step rationales before outputting the answer to effectively communicate the task to the model. However, as sequences become longer, we find that current models struggle to keep track of token positions. To address this issue, we interleave output tokens with markup tokens that act as explicit positional and counting symbols. Our findings show how these two complementary approaches enable remarkable sequence extrapolation and highlight a limitation of current architectures to effectively generalize without explicit surface form guidance. Code available at //github.com/MirelleB/induced-rationales-markup-tokens
Although large pre-trained language models have achieved great success in many NLP tasks, it has been shown that they reflect human biases from their pre-training corpora. This bias may lead to undesirable outcomes when these models are applied in real-world settings. In this paper, we investigate the bias present in monolingual BERT models across a diverse set of languages (English, Greek, and Persian). While recent research has mostly focused on gender-related biases, we analyze religious and ethnic biases as well and propose a template-based method to measure any kind of bias, based on sentence pseudo-likelihood, that can handle morphologically complex languages with gender-based adjective declensions. We analyze each monolingual model via this method and visualize cultural similarities and differences across different dimensions of bias. Ultimately, we conclude that current methods of probing for bias are highly language-dependent, necessitating cultural insights regarding the unique ways bias is expressed in each language and culture (e.g. through coded language, synecdoche, and other similar linguistic concepts). We also hypothesize that higher measured social biases in the non-English BERT models correlate with user-generated content in their training.
We propose a new data-driven method to learn the dynamics of an unknown hyperbolic system of conservation laws using deep neural networks. Inspired by classical methods in numerical conservation laws, we develop a new conservative form network (CFN) in which the network learns the flux function of the unknown system. Our numerical examples demonstrate that the CFN yields significantly better prediction accuracy than what is obtained using a standard non-conservative form network, even when it is enhanced with constraints to promote conservation. In particular, solutions obtained using the CFN consistently capture the correct shock propagation speed without introducing non-physical oscillations into the solution. They are furthermore robust to noisy and sparse observation environments.
Electronic shot counters allow armourers to perform preventive and predictive maintenance based on quantitative measurements, improving reliability, reducing the frequency of accidents, and reducing maintenance costs. To answer a market pressure for both low lead time to market and increased customisation, we aim to solve the shot detection and shot counting problem in a generic way through machine learning. In this study, we describe a method allowing one to construct a dataset with minimal labelling effort by only requiring the total number of shots fired in a time series. To our knowledge, this is the first study to propose a technique, based on learning from label proportions, that is able to exploit these weak labels to derive an instance-level classifier able to solve the counting problem and the more general discrimination problem. We also show that this technique can be deployed in heavily constrained microcontrollers while still providing hard real-time (<100ms) inference. We evaluate our technique against a state-of-the-art unsupervised algorithm and show a sizeable improvement, suggesting that the information from the weak labels is successfully leveraged. Finally, we evaluate our technique against human-generated state-of-the-art algorithms and show that it provides comparable performance and significantly outperforms them in some offline and real-world benchmarks.
We present a new construction of high dimensional expanders based on covering spaces of simplicial complexes. High dimensional expanders (HDXs) are hypergraph analogues of expander graphs. They have many uses in theoretical computer science, but unfortunately only few constructions are known which have arbitrarily small local spectral expansion. We give a randomized algorithm that takes as input a high dimensional expander $X$ (satisfying some mild assumptions). It outputs a sub-complex $Y \subseteq X$ that is a high dimensional expander and has infinitely many simplicial covers. These covers form new families of bounded-degree high dimensional expanders. The sub-complex $Y$ inherits $X$'s underlying graph and its links are sparsifications of the links of $X$. When the size of the links of $X$ is $O(\log |X|)$, this algorithm can be made deterministic. Our algorithm is based on the groups and generating sets discovered by Lubotzky, Samuels and Vishne (2005), that were used to construct the first discovered high dimensional expanders. We show these groups give rise to many more ``randomized'' high dimensional expanders. In addition, our techniques also give a random sparsification algorithm for high dimensional expanders, that maintains its local spectral properties. This may be of independent interest.
Improving open-ended learning capabilities is a promising approach to enable robots to face the unbounded complexity of the real-world. Among existing methods, the ability of Quality-Diversity algorithms to generate large collections of diverse and high-performing skills is instrumental in this context. However, most of those algorithms rely on a hand-coded behavioural descriptor to characterise the diversity, hence requiring prior knowledge about the considered tasks. In this work, we propose an additional analysis of Autonomous Robots Realising their Abilities; a Quality-Diversity algorithm that autonomously finds behavioural characterisations. We evaluate this approach on a simulated robotic environment, where the robot has to autonomously discover its abilities from its full-state trajectories. All algorithms were applied to three tasks: navigation, moving forward with a high velocity, and performing half-rolls. The experimental results show that the algorithm under study discovers autonomously collections of solutions that are diverse with respect to all tasks. More specifically, the analysed approach autonomously finds policies that make the robot move to diverse positions, but also utilise its legs in diverse ways, and even perform half-rolls.
Transformers have achieved superior performances in many tasks in natural language processing and computer vision, which also intrigues great interests in the time series community. Among multiple advantages of transformers, the ability to capture long-range dependencies and interactions is especially attractive for time series modeling, leading to exciting progress in various time series applications. In this paper, we systematically review transformer schemes for time series modeling by highlighting their strengths as well as limitations through a new taxonomy to summarize existing time series transformers in two perspectives. From the perspective of network modifications, we summarize the adaptations of module level and architecture level of the time series transformers. From the perspective of applications, we categorize time series transformers based on common tasks including forecasting, anomaly detection, and classification. Empirically, we perform robust analysis, model size analysis, and seasonal-trend decomposition analysis to study how Transformers perform in time series. Finally, we discuss and suggest future directions to provide useful research guidance. To the best of our knowledge, this paper is the first work to comprehensively and systematically summarize the recent advances of Transformers for modeling time series data. We hope this survey will ignite further research interests in time series Transformers.
Since hardware resources are limited, the objective of training deep learning models is typically to maximize accuracy subject to the time and memory constraints of training and inference. We study the impact of model size in this setting, focusing on Transformer models for NLP tasks that are limited by compute: self-supervised pretraining and high-resource machine translation. We first show that even though smaller Transformer models execute faster per iteration, wider and deeper models converge in significantly fewer steps. Moreover, this acceleration in convergence typically outpaces the additional computational overhead of using larger models. Therefore, the most compute-efficient training strategy is to counterintuitively train extremely large models but stop after a small number of iterations. This leads to an apparent trade-off between the training efficiency of large Transformer models and the inference efficiency of small Transformer models. However, we show that large models are more robust to compression techniques such as quantization and pruning than small models. Consequently, one can get the best of both worlds: heavily compressed, large models achieve higher accuracy than lightly compressed, small models.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191