Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userREX: Revisiting Budgeted Training with an Improved Schedule
87
0
0.0
(
0
)
Added by
John Chen
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Deep learning practitioners often operate on a computational and monetary budget. Thus, it is critical to design optimization algorithms that perform well under any budget. The linear learning rate schedule is considered the best budget-aware schedule, as it outperforms most other schedules in the low budget regime. On the other hand, learning rate schedules -- such as the texttt{30-60-90} step schedule -- are known to achieve high performance when the model can be trained for many epochs. Yet, it is often not known a priori whether ones budget will be large or small; thus, the optimal choice of learning rate schedule is made on a case-by-case basis. In this paper, we frame the learning rate schedule selection problem as a combination of $i)$ selecting a profile (i.e., the continuous function that models the learning rate schedule), and $ii)$ choosing a sampling rate (i.e., how frequently the learning rate is updated/sampled from this profile). We propose a novel profile and sampling rate combination called the Reflected Exponential (REX) schedule, which we evaluate across seven different experimental settings with both SGD and Adam optimizers. REX outperforms the linear schedule in the low budget regime, while matching or exceeding the performance of several state-of-the-art learning rate schedules (linear, step, exponential, cosine, step decay on plateau, and OneCycle) in both high and low budget regimes. Furthermore, REX requires no added computation, storage, or hyperparameters.
rate research
Read More
State-of-the-art generic low-precision training algorithms use a mix of 16-bit and 32-bit precision, creating the folklore that 16-bit hardware compute units alone are not enough to maximize model accuracy. As a result, deep learning accelerators are forced to support both 16-bit and 32-bit floating-point units (FPUs), which is more costly than only using 16-bit FPUs for hardware design. We ask: can we train deep learning models only with 16-bit floating-point units, while still matching the model accuracy attained by 32-bit training? Towards this end, we study 16-bit-FPU training on the widely adopted BFloat16 unit. While these units conventionally use nearest rounding to cast output to 16-bit precision, we show that nearest rounding for model weight updates often cancels small updates, which degrades the convergence and model accuracy. Motivated by this, we study two simple techniques well-established in numerical analysis, stochastic rounding and Kahan summation, to remedy the model accuracy degradation in 16-bit-FPU training. We demonstrate that these two techniques can enable up to 7% absolute validation accuracy gain in 16-bit-FPU training. This leads to 0.1% lower to 0.2% higher validation accuracy compared to 32-bit training across seven deep learning applications.
Momentum is a widely used technique for gradient-based optimizers in deep learning. In this paper, we propose a decaying momentum (textsc{Demon}) rule. We conduct the first large-scale empirical analysis of momentum decay methods for modern neural network optimization, in addition to the most popular learning rate decay schedules. Across 28 relevant combinations of models, epochs, datasets, and optimizers, textsc{Demon} achieves the highest number of Top-1 and Top-3 finishes at 39% and 85% respectively, almost doubling the second-placed learning rate cosine schedule at 17% and 60%, respectively. textsc{Demon} also outperforms other widely used schedulers including, but not limited to, the learning rate step schedule, linear schedule, OneCycle schedule, and exponential schedule. Compared with the widely used learning rate step schedule, textsc{Demon} is observed to be less sensitive to parameter tuning, which is critical to training neural networks in practice. Results are demonstrated across a variety of settings and architectures, including image classification, generative models, and language models. textsc{Demon} is easy to implement, requires no additional tuning, and incurs almost no extra computational overhead compared to the vanilla counterparts. Code is readily available.
Self-training is one of the earliest and simplest semi-supervised methods. The key idea is to augment the original labeled dataset with unlabeled data paired with the models prediction (i.e. the pseudo-parallel data). While self-training has been extensively studied on classification problems, in complex sequence generation tasks (e.g. machine translation) it is still unclear how self-training works due to the compositionality of the target space. In this work, we first empirically show that self-training is able to decently improve the supervised baseline on neural sequence generation tasks. Through careful examination of the performance gains, we find that the perturbation on the hidden states (i.e. dropout) is critical for self-training to benefit from the pseudo-parallel data, which acts as a regularizer and forces the model to yield close predictions for similar unlabeled inputs. Such effect helps the model correct some incorrect predictions on unlabeled data. To further encourage this mechanism, we propose to inject noise to the input space, resulting in a noisy version of self-training. Empirical study on standard machine translation and text summarization benchmarks shows that noisy self-training is able to effectively utilize unlabeled data and improve the performance of the supervised baseline by a large margin.
Adversarial training has shown impressive success in learning bilingual dictionary without any parallel data by mapping monolingual embeddings to a shared space. However, recent work has shown superior performance for non-adversarial methods in more challenging language pairs. In this work, we revisit adversarial autoencoder for unsupervised word translation and propose two novel extensions to it that yield more stable training and improved results. Our method includes regularization terms to enforce cycle consistency and input reconstruction, and puts the target encoders as an adversary against the corresponding discriminator. Extensive experimentations with European, non-European and low-resource languages show that our method is more robust and achieves better performance than recently proposed adversarial and non-adversarial approaches.
A remarkable recent discovery in machine learning has been that deep neural networks can achieve impressive performance (in terms of both lower training error and higher generalization capacity) in the regime where they are massively over-parameterized. Consequently, over the past year, the community has devoted growing interest in analyzing optimization and generalization properties of over-parameterized networks, and several breakthrough works have led to important theoretical progress. However, the majority of existing work only applies to supervised learning scenarios and hence are limited to settings such as classification and regression. In contrast, the role of over-parameterization in the unsupervised setting has gained far less attention. In this paper, we study the gradient dynamics of two-layer over-parameterized autoencoders with ReLU activation. We make very few assumptions about the given training dataset (other than mild non-degeneracy conditions). Starting from a randomly initialized autoencoder network, we rigorously prove the linear convergence of gradient descent in two learning regimes, namely: (i) the weakly-trained regime where only the encoder is trained, and (ii) the jointly-trained regime where both the encoder and the decoder are trained. Our results indicate the considerable benefits of joint training over weak training for finding global optima, achieving a dramatic decrease in the required level of over-parameterization. We also analyze the case of weight-tied autoencoders (which is a commonly used architectural choice in practical settings) and prove that in the over-parameterized setting, training such networks from randomly initialized points leads to certain unexpected degeneracies.
suggested questions
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
