Training deep learning models can be computationally expensive. Prior works have shown that increasing the batch size can potentially lead to better overall throughput. However, the batch size is frequently limited by the accelerator memory capacity due to the activations/feature maps stored for the training backward pass, as larger batch sizes require larger feature maps to be stored. Transformer-based models, which have recently seen a surge in popularity due to their good performance and applicability to a variety of tasks, have a similar problem. To remedy this issue, we propose Tempo, a new approach to efficiently use accelerator (e.g., GPU) memory resources for training Transformer-based models. Our approach provides drop-in replacements for the GELU, LayerNorm, and Attention layers, reducing the memory usage and ultimately leading to more efficient training. We implement Tempo and evaluate the throughput, memory usage, and accuracy/loss on the BERT Large pre-training task. We demonstrate that Tempo enables up to 2 higher batch sizes and 16% higher training throughput over the state-of-the-art baseline. We also evaluate Tempo on GPT2 and RoBERTa models, showing 19% and 26% speedup over the baseline.

In this paper, we explore the limits of Microsoft Floating Point (MSFP), a new class of datatypes developed for production cloud-scale inferencing on custom hardware. Through the co-evolution of hardware design and algorithms, MSFP achieves accuracy comparable to or better than industry standards Bfloat16 and INT8 at 3x and 4x lower cost, respectively. MSFP incurs negligible impact to accuracy (<1%), requires no changes to the model topology, and is integrated with a mature cloud production pipeline. MSFP supports various classes of deep learning models including CNNs, RNNs, and Transformers without modification. Finally, we characterize the accuracy and implementation of MSFP and demonstrate its efficacy on a number of production scenarios, including models that power major online scenarios such as web search, question-answering, and image classification.

Spatio-temporal representational learning has been widely adopted in various fields such as action recognition, video object segmentation, and action anticipation.Previous spatio-temporal representational learning approaches primarily employ ConvNets or sequential models, e.g., LSTM, to learn the intra-frame and inter-frame features. Recently, Transformer models have successfully dominated the study of natural language processing (NLP), image classification, etc. However, the pure-Transformer based spatio-temporal learning can be prohibitively costly on memory and computation to extract fine-grained features from a tiny patch. To tackle the training difficulty and enhance the spatio-temporal learning, we construct a shifted chunk Transformer with pure self-attention blocks. Leveraging the recent efficient Transformer design in NLP, this shifted chunk Transformer can learn hierarchical spatio-temporal features from a local tiny patch to a global videoclip. Our shifted self-attention can also effectively model complicated inter-frame variances. Furthermore, we build a clip encoder based on Transformer to model long-term temporal dependencies. We conduct thorough ablation studies to validate each component and hyper-parameters in our shifted chunk Transformer, and it outperforms previous state-of-the-art approaches on Kinetics-400, Kinetics-600,UCF101, and HMDB51.

There are two types of deep generative models: explicit and implicit. The former defines an explicit density form that allows likelihood inference; while the latter targets a flexible transformation from random noise to generated samples. While the two classes of generative models have shown great power in many applications, both of them, when used alone, suffer from respective limitations and drawbacks. To take full advantages of both models and enable mutual compensation, we propose a novel joint training framework that bridges an explicit (unnormalized) density estimator and an implicit sample generator via Stein discrepancy. We show that our method 1) induces novel mutual regularization via kernel Sobolev norm penalization and Moreau-Yosida regularization, and 2) stabilizes the training dynamics. Empirically, we demonstrate that proposed method can facilitate the density estimator to more accurately identify data modes and guide the generator to output higher-quality samples, comparing with training a single counterpart. The new approach also shows promising results when the training samples are contaminated or limited.

We propose a novel training method that integrates rules into deep learning, in a way the strengths of the rules are controllable at inference. Deep Neural Networks with Controllable Rule Representations (DeepCTRL) incorporates a rule encoder into the model coupled with a rule-based objective, enabling a shared representation for decision making. DeepCTRL is agnostic to data type and model architecture. It can be applied to any kind of rule defined for inputs and outputs. The key aspect of DeepCTRL is that it does not require retraining to adapt the rule strength -- at inference, the user can adjust it based on the desired operation point on accuracy vs. rule verification ratio. In real-world domains where incorporating rules is critical -- such as Physics, Retail and Healthcare -- we show the effectiveness of DeepCTRL in teaching rules for deep learning. DeepCTRL improves the trust and reliability of the trained models by significantly increasing their rule verification ratio, while also providing accuracy gains at downstream tasks. Additionally, DeepCTRL enables novel use cases such as hypothesis testing of the rules on data samples, and unsupervised adaptation based on shared rules between datasets.

We present QLoRA, an efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB GPU while preserving full 16-bit finetuning task performance. QLoRA backpropagates gradients through a frozen, 4-bit quantized pretrained language model into Low Rank Adapters~(LoRA). Our best model family, which we name Guanaco, outperforms all previous openly released models on the Vicuna benchmark, reaching 99.3% of the performance level of ChatGPT while only requiring 24 hours of finetuning on a single GPU. QLoRA introduces a number of innovations to save memory without sacrificing performance: (a) 4-bit NormalFloat (NF4), a new data type that is information-theoretically optimal for normally distributed weights (b) Double Quantization to reduce the average memory footprint by quantizing the quantization constants, and (c) Paged Optimziers to manage memory spikes. We use QLoRA to finetune more than 1,000 models, providing a detailed analysis of instruction following and chatbot performance across 8 instruction datasets, multiple model types (LLaMA, T5), and model scales that would be infeasible to run with regular finetuning (e.g.

Recently, Transformer has become a prevailing deep architecture for solving vehicle routing problems (VRPs). However, it is less effective in learning improvement models for VRP because its positional encoding (PE) method is not suitable in representing VRP solutions. This paper presents a novel Dual-Aspect Collaborative Transformer (DACT) to learn embeddings for the node and positional features separately, instead of fusing them together as done in existing ones, so as to avoid potential noises and incompatible correlations. Moreover, the positional features are embedded through a novel cyclic positional encoding (CPE) method to allow Transformer to effectively capture the circularity and symmetry of VRP solutions (i.e., cyclic sequences). We train DACT using Proximal Policy Optimization and design a curriculum learning strategy for better sample efficiency. We apply DACT to solve the traveling salesman problem (TSP) and capacitated vehicle routing problem (CVRP). Results show that our DACT outperforms existing Transformer based improvement models, and exhibits much better generalization performance across different problem sizes on synthetic and benchmark instances, respectively.

Within months of birth, children develop meaningful expectations about the world around them. How much of this early knowledge can be explained through generic learning mechanisms applied to sensory data, and how much of it requires more substantive innate inductive biases? Addressing this fundamental question in its full generality is currently infeasible, but we can hope to make real progress in more narrowly defined domains, such as the development of high-level visual categories, thanks to improvements in data collecting technology and recent progress in deep learning. In this paper, our goal is precisely to achieve such progress by utilizing modern self-supervised deep learning methods and a recent longitudinal, egocentric video dataset recorded from the perspective of three young children (Sullivan et al., 2020). Our results demonstrate the emergence of powerful, high-level visual representations from developmentally realistic natural videos using generic self-supervised learning objectives.

Structured latent variables allow incorporating meaningful prior knowledge into deep learning models. However, learning with such variables remains challenging because of their discrete nature. Nowadays, the standard learning approach is to define a latent variable as a perturbed algorithm output and to use a differentiable surrogate for training. In general, the surrogate puts additional constraints on the model and inevitably leads to biased gradients. To alleviate these shortcomings, we extend the Gumbel-Max trick to define distributions over structured domains. We avoid the differentiable surrogates by leveraging the score function estimators for optimization. In particular, we highlight a family of recursive algorithms with a common feature we call stochastic invariant. The feature allows us to construct reliable gradient estimates and control variates without additional constraints on the model. In our experiments, we consider various structured latent variable models and achieve results competitive with relaxation-based counterparts.

Two key challenges facing modern deep learning is mitigating deep networks vulnerability to adversarial attacks, and understanding deep learning's generalization capabilities. Towards the first issue, many defense strategies have been developed, with the most common being Adversarial Training (AT). Towards the second challenge, one of the dominant theories that has emerged is the Neural Tangent Kernel (NTK) -- a characterization of neural network behavior in the infinite-width limit. In this limit, the kernel is frozen and the underlying feature map is fixed. In finite-widths however, there is evidence that feature learning happens at the earlier stages of the training (kernel learning) before a second phase where the kernel remains fixed (lazy training).