Large Language Models (LLMs) are reshaping the research landscape in artificial intelligence, particularly as model parameters scale up significantly, unlocking remarkable capabilities across various domains. Nevertheless, the scalability of model parameters faces constraints due to limitations in GPU memory and computational speed. To address these constraints, various weight compression methods have emerged, such as Pruning and Quantization. Given the low-rank nature of weight matrices in language models, the reduction of weights through matrix decomposition undoubtedly holds significant potential and promise. In this paper, drawing upon the intrinsic structure of LLMs, we propose a novel approach termed Data-free Joint Rank-k Approximation for compressing the parameter matrices. Significantly, our method is characterized by without necessitating additional involvement of any corpus, while simultaneously preserving orthogonality in conjunction with pruning and quantization methods. We achieve a model pruning of 80% parameters while retaining 93.43% of the original performance without any calibration data. Additionally, we explore the fundamental properties of the weight matrix of LLMs undergone Rank-k Approximation and conduct comprehensive experiments to elucidate our hypothesis.

Memory-efficient continuous Sign Language Translation is a significant challenge for the development of assisted technologies with real-time applicability for the deaf. In this work, we introduce a paradigm of designing recurrent deep networks whereby the output of the recurrent layer is derived from appropriate arguments from nonparametric statistics. A novel variational Bayesian sequence-to-sequence network architecture is proposed that consists of a) a full Gaussian posterior distribution for data-driven memory compression and b) a nonparametric Indian Buffet Process prior for regularization applied on the Gated Recurrent Unit non-gate weights. We dub our approach Stick-Breaking Recurrent network and show that it can achieve a substantial weight compression without diminishing modeling performance.

We introduce a technique that compresses deep neural networks both during and after training by constraining the total number of weights updated during backpropagation to those with the highest total gradients. The remaining weights are forgotten and their initial value is regenerated at every access to avoid storing them in memory. This dramatically reduces the number of off-chip memory accesses during both training and inference, a key component of the energy needs of DNN accelerators. By ensuring that the total weight diffusion remains close to that of baseline unpruned SGD, networks pruned using DropBack are able to maintain high accuracy across network architectures. We observe weight compression of 25x with LeNet-300-100 on MNIST while maintaining accuracy. On CIFAR-10, we see an approximately 5x weight compression on 3 models: an already 9x-reduced VGG-16, Densenet, and WRN-28-10 - all with zero or negligible accuracy loss. On Densenet and WRN, which are particularly challenging to compress, Both Densenet and WRN improve on the state of the art, achieving higher compression with better accuracy than prior pruning techniques.