上下文压缩

Quantization of the weights and activations is one of the main methods to reduce the computational footprint of Deep Neural Networks (DNNs) training. Current methods enable 4-bit quantization of the forward phase. However, this constitutes only a third of the training process. Reducing the computational footprint of the entire training process requires the quantization of the neural gradients, i.e., the loss gradients with respect to the outputs of intermediate neural layers. Previous works separately showed that accurate 4-bit quantization of the neural gradients needs to (1) be unbiased and (2) have a log scale. However, no previous work aimed to combine both ideas, as we do in this work. Specifically, we examine the importance of having unbiased quantization in quantized neural network training, where to maintain it, and how to combine it with logarithmic quantization. Based on this, we suggest a $\textit{logarithmic unbiased quantization}$ (LUQ) method to quantize all both the forward and backward phase to 4-bit, achieving state-of-the-art results in 4-bit training without overhead. For example, in ResNet50 on ImageNet, we achieved a degradation of 1.1 %. We further improve this to degradation of only 0.32 % after three epochs of high precision fine-tunining, combined with a variance reduction method---where both these methods add overhead comparable to previously suggested methods. A reference implementation is supplied in the supplementary material.

Achieving structured, data-free sparsity of deep neural networks (DNNs) remains an open area of research. In this work, we address the challenge of pruning filters without access to the original training set or loss function. We propose the discriminative filters hypothesis, that well-trained models possess discriminative filters, and any non-discriminative filters can be pruned without impacting the predictive performance of the classifier. Based on this hypothesis, we propose a new paradigm for pruning neural networks: distributional pruning, wherein we only require access to the distributions that generated the original datasets. Our approach to solving the problem of formalising and quantifying the discriminating ability of filters is through the total variation (TV) distance between the class-conditional distributions of the filter outputs. We present empirical results that, using this definition of discriminability, support our hypothesis on a variety of datasets and architectures. Next, we define the LDIFF score, a heuristic to quantify the extent to which a layer possesses a mixture of discriminative and non-discriminative filters. We empirically demonstrate that the LDIFF score is indicative of the performance of random pruning for a given layer, and thereby indicates the extent to which a layer may be pruned. Our main contribution is a novel one-shot pruning algorithm, called TVSPrune, that identifies non-discriminative filters for pruning. We extend this algorithm to IterTVSPrune, wherein we iteratively apply TVSPrune, thereby enabling us to achieve greater sparsity. Last, we demonstrate the efficacy of the TVSPrune on a variety of datasets, and show that in some cases, we can prune up to 60% of parameters with only a 2% loss of accuracy without any fine-tuning of the model, beating the nearest baseline by almost 10%.

Tensor decompositions have been successfully applied to compress neural networks. The compression algorithms using tensor decompositions commonly minimize the approximation error on the weights. Recent work assumes the approximation error on the weights is a proxy for the performance of the model to compress multiple layers and fine-tune the compressed model. Surprisingly, little research has systematically evaluated which approximation errors can be used to make choices regarding the layer, tensor decomposition method, and level of compression. To close this gap, we perform an experimental study to test if this assumption holds across different layers and types of decompositions, and what the effect of fine-tuning is. We include the approximation error on the features resulting from a compressed layer in our analysis to test if this provides a better proxy, as it explicitly takes the data into account. We find the approximation error on the weights has a positive correlation with the performance error, before as well as after fine-tuning. Basing the approximation error on the features does not improve the correlation significantly. While scaling the approximation error commonly is used to account for the different sizes of layers, the average correlation across layers is smaller than across all choices (i.e. layers, decompositions, and level of compression) before fine-tuning. When calculating the correlation across the different decompositions, the average rank correlation is larger than across all choices. This means multiple decompositions can be considered for compression and the approximation error can be used to choose between them.

Modern, multi-branched neural network architectures often possess complex interconnections between layers, which we call coupled channels (CCs). Structured pruning of CCs in these multi-branch networks is an under-researched problem, as most existing works are typically designed for pruning single-branch models like VGG-nets. While these methods yield accurate subnetworks, the improvements in inference times when applied to multi-branch networks are comparatively modest, as these methods do not prune CCs, which we observe contribute significantly to inference time. For instance, layers with CCs as input or output take more than 66% of the inference time in ResNet-50. Moreover, pruning in the data-free regime, where data is not used for pruning, is gaining traction owing to privacy concerns and computational costs associated with fine-tuning. Motivated by this, we study the problem of pruning CCs in the data-free regime. To facilitate the development of algorithms to prune CCs, we define Data Flow Couplings (DFCs) to enumerate the layers that constitute coupled connections and the associated transformation. Additionally, saliencies for pruning CCs cannot be gauged in isolation, as there may be discrepancies among the layerwise importance of CCs using conventional scoring strategies. This necessitates finding grouped saliencies to gauge the importance of all corresponding coupled elements in a network. We thus propose the Backwards Graph-based Saliency Computation (BGSC) algorithm, a data-free method that computes saliencies by estimating an upper bound to the reconstruction error of intermediate layers; we call this pruning strategy Data Flow driven Pruning of Coupled channels (DFPC). Finally, we show the efficacy of DFPC for models trained on standard datasets. Since we pruned coupled channels, we achieve up to 1.66x improvements in inference time for ResNet-101 trained on CIFAR-10 with a 5% accuracy drop without fine-tuning. With access to the ImageNet training set, we achieve significant improvements over the data-free method and see an improvement of at least 47.1% in speedup for a 2.3% accuracy drop for ResNet-50 against our baselines.

Deep learning video compression outperforms its hand-craft counterparts with enhanced flexibility and capacity. One key component of the learned video codec is the autoregressive entropy model conditioned on spatial and temporal priors. Operating autoregressive on raster scanning order naively treats the context as unidirectional. This is neither efficient nor optimal, considering that conditional information probably locates at the end of the sequence. We thus introduce an entropy model based on a masked image modeling transformer (MIMT) to learn the spatial-temporal dependencies. Video frames are first encoded into sequences of tokens and then processed with the transformer encoder as priors. The transformer decoder learns the probability mass functions (PMFs) \emph{conditioned} on the priors and masked inputs. Then it is capable of selecting optimal decoding orders without a fixed direction. During training, MIMT aims to predict the PMFs of randomly masked tokens by attending to tokens in all directions. This allows MIMT to capture the temporal dependencies from encoded priors and the spatial dependencies from the unmasked tokens, i.e., decoded tokens. At inference time, the model begins with generating PMFs of all masked tokens in parallel and then decodes the frame iteratively from the previously-selected decoded tokens (i.e., with high confidence). In addition, we improve the overall performance with more techniques, e.g., manifold conditional priors accumulating a long range of information, shifted window attention to reduce complexity. Extensive experiments demonstrate the proposed MIMT framework equipped with the new transformer entropy model achieves state-of-the-art performance on HEVC, UVG, and MCL-JCV datasets, generally outperforming the VVC in terms of PSNR and SSIM.

In recent years, deep network pruning has attracted significant attention in order to enable the rapid deployment of AI into small devices with computation and memory constraints. Pruning is often achieved by dropping redundant weights, neurons, or layers of a deep network while attempting to retain a comparable test performance. Many deep pruning algorithms have been proposed with impressive empirical success. However, existing approaches lack a quantifiable measure to estimate the compressibility of a sub-network during each pruning iteration and thus may under-prune or over-prune the model. In this work, we propose PQ Index (PQI) to measure the potential compressibility of deep neural networks and use this to develop a Sparsity-informed Adaptive Pruning (SAP) algorithm. Our extensive experiments corroborate the hypothesis that for a generic pruning procedure, PQI decreases first when a large model is being effectively regularized and then increases when its compressibility reaches a limit that appears to correspond to the beginning of underfitting. Subsequently, PQI decreases again when the model collapse and significant deterioration in the performance of the model start to occur. Additionally, our experiments demonstrate that the proposed adaptive pruning algorithm with proper choice of hyper-parameters is superior to the iterative pruning algorithms such as the lottery ticket-based pruning methods, in terms of both compression efficiency and robustness.

In-Context Learning (ICL), which formulates target tasks as prompt completion conditioned on in-context demonstrations, has become the prevailing utilization of LLMs. In this paper, we first disclose an actual predicament for this typical usage that it can not scale up with training data due to context length restriction. Besides, existing works have shown that ICL also suffers from various biases and requires delicate calibration treatment. To address both challenges, we advocate a simple and effective solution, $k$NN Prompting, which first queries LLM with training data for distributed representations, then predicts test instances by simply referring to nearest neighbors. We conduct comprehensive experiments to demonstrate its two-fold superiority: 1) Calibration-Free: $k$NN Prompting does not directly align LLM output distribution with task-specific label space, instead leverages such distribution to align test and training instances. It significantly outperforms state-of-the-art calibration-based methods under comparable few-shot scenario. 2) Beyond-Context: $k$NN Prompting can further scale up effectively with as many training data as are available, continually bringing substantial improvements. The scaling trend holds across 10 orders of magnitude ranging from 2 shots to 1024 shots as well as different LLMs scales ranging from 0.8B to 30B. It successfully bridges data scaling into model scaling, and brings new potentials for the gradient-free paradigm of LLM deployment. Code is publicly available at https://github.com/BenfengXu/KNNPrompting

Knowledge-intensive tasks, such as open-domain question answering (QA), require access to a large amount of world or domain knowledge. A common approach for knowledge-intensive tasks is to employ a retrieve-then-read pipeline that first retrieves a handful of relevant contextual documents from an external corpus such as Wikipedia and then predicts an answer conditioned on the retrieved documents. In this paper, we present a novel perspective for solving knowledge-intensive tasks by replacing document retrievers with large language model generators. We call our method generate-then-read (GenRead), which first prompts a large language model to generate contextual documents based on a given question, and then reads the generated documents to produce the final answer. Furthermore, we propose a novel clustering-based prompting method that selects distinct prompts, in order to generate diverse documents that cover different perspectives, leading to better recall over acceptable answers. We conduct extensive experiments on three different knowledge-intensive tasks, including open-domain QA, fact checking, and dialogue system. Notably, GenRead achieves 71.6 and 54.4 exact match scores on TriviaQA and WebQ, significantly outperforming the state-of-the-art retrieve-then-read pipeline DPR-FiD by +4.0 and +3.9, without retrieving any documents from any external knowledge source. Lastly, we demonstrate the model performance can be further improved by combining retrieval and generation. Our code and generated documents can be found at https://github.com/wyu97/GenRead.

Biological neural networks are capable of recruiting different sets of neurons to encode different memories. However, when training artificial neural networks on a set of tasks, typically, no mechanism is employed for selectively producing anything analogous to these neuronal ensembles. Further, artificial neural networks suffer from catastrophic forgetting, where the network's performance rapidly deteriorates as tasks are learned sequentially. By contrast, sequential learning is possible for a range of biological organisms. We introduce Learned Context Dependent Gating (LXDG), a method to flexibly allocate and recall `artificial neuronal ensembles', using a particular network structure and a new set of regularization terms. Activities in the hidden layers of the network are modulated by gates, which are dynamically produced during training. The gates are outputs of networks themselves, trained with a sigmoid output activation. The regularization terms we have introduced correspond to properties exhibited by biological neuronal ensembles. The first term penalizes low gate sparsity, ensuring that only a specified fraction of the network is used. The second term ensures that previously learned gates are recalled when the network is presented with input from previously learned tasks. Finally, there is a regularization term responsible for ensuring that new tasks are encoded in gates that are as orthogonal as possible from previously used ones. We demonstrate the ability of this method to alleviate catastrophic forgetting on continual learning benchmarks. When the new regularization terms are included in the model along with Elastic Weight Consolidation (EWC) it achieves better performance on the benchmark `permuted MNIST' than with EWC alone. The benchmark `rotated MNIST' demonstrates how similar tasks recruit similar neurons to the artificial neuronal ensemble.

Soft threshold pruning is among the cutting-edge pruning methods with state-of-the-art performance. However, previous methods either perform aimless searching on the threshold scheduler or simply set the threshold trainable, lacking theoretical explanation from a unified perspective. In this work, we reformulate soft threshold pruning as an implicit optimization problem solved using the Iterative Shrinkage-Thresholding Algorithm (ISTA), a classic method from the fields of sparse recovery and compressed sensing. Under this theoretical framework, all threshold tuning strategies proposed in previous studies of soft threshold pruning are concluded as different styles of tuning $L_1$-regularization term. We further derive an optimal threshold scheduler through an in-depth study of threshold scheduling based on our framework. This scheduler keeps $L_1$-regularization coefficient stable, implying a time-invariant objective function from the perspective of optimization. In principle, the derived pruning algorithm could sparsify any mathematical model trained via SGD. We conduct extensive experiments and verify its state-of-the-art performance on both Artificial Neural Networks (ResNet-50 and MobileNet-V1) and Spiking Neural Networks (SEW ResNet-18) on ImageNet datasets. On the basis of this framework, we derive a family of pruning methods, including sparsify-during-training, early pruning, and pruning at initialization. The code is available at https://github.com/Yanqi-Chen/LATS.

Deep neural networks (DNNs) often have to be compressed, via pruning and/or quantization, before they can be deployed in practical settings. In this work we propose a new compression-aware minimizer dubbed CrAM that modifies the optimization step in a principled way, in order to produce models whose local loss behavior is stable under compression operations such as pruning. Thus, dense models trained via CrAM should be compressible post-training, in a single step, without significant accuracy loss. Experimental results on standard benchmarks, such as residual networks for ImageNet classification and BERT models for language modelling, show that CrAM produces dense models that can be more accurate than the standard SGD/Adam-based baselines, but which are stable under weight pruning: specifically, we can prune models in one-shot to 70-80% sparsity with almost no accuracy loss, and to 90% with reasonable (∼ 1%) accuracy loss, which is competitive with gradual compression methods. Additionally, CrAM can produce sparse models which perform well for transfer learning, and it also works for semi-structured 2:4 pruning patterns supported by GPU hardware. The code for reproducing the results is available at: https://github.com/IST-DASLab/CrAM .

Training deep reinforcement learning (DRL) models usually requires high computation costs. Therefore, compressing DRL models possesses immense potential for training acceleration and model deployment. However, existing methods that generate small models mainly adopt the knowledge distillation-based approach by iteratively training a dense network. As a result, the training process still demands massive computing resources. Indeed, sparse training from scratch in DRL has not been well explored and is particularly challenging due to non-stationarity in bootstrap training. In this work, we propose a novel sparse DRL training framework, “the Rigged Reinforcement Learning Lottery” (RLx2), which builds upon gradient-based topology evolution and is capable of training a sparse DRL model based entirely on a sparse network. Specifically, RLx2 introduces a novel multi-step TD target mechanism with a dynamic-capacity replay buffer to achieve robust value learning and efficient topology exploration in sparse models. It also reaches state-of-the-art sparse training performance in several tasks, showing $7.5\times$-$20\times$ model compression with less than $3\%$ performance degradation and up to $20\times$ and $50\times$ FLOPs reduction for training and inference, respectively.

Neural sequence models, especially transformers, exhibit a remarkable capacity for in-context learning. They can construct new predictors from sequences of labeled examples $(x, f(x))$ presented in the input without further parameter updates. We investigate the hypothesis that transformer-based in-context learners implement standard learning algorithms implicitly, by encoding context-specific parametric models in their hidden representations, and updating these implicit models as new examples appear in the context. Using linear regression as a model problem, we offer three sources of evidence for this hypothesis. First, we prove by construction that transformers can implement learning algorithms for linear models based on gradient descent and closed-form computation of regression parameters. Second, we show that trained in-context learners closely match the predictors computed by gradient descent, ridge regression, and exact least-squares regression, transitioning between different predictors as transformer depth and dataset noise vary. Third, we present preliminary evidence that in-context learners share algorithmic features with these predictors: learners' late layers encode weight vectors and moment matrices. These results suggest that in-context learning is understandable in algorithmic terms, and that (at least in the linear case) learners may work by rediscovering standard estimation algorithms.