Brain tumors are one of the most common diseases that lead to early death if not diagnosed at an early stage. Traditional diagnostic approaches are extremely time-consuming and prone to errors. In this context, computer vision-based approaches have emerged as an effective tool for accurate brain tumor classification. While some of the existing solutions demonstrate noteworthy accuracy, the models become infeasible to deploy in areas where computational resources are limited. This research addresses the need for accurate and fast classification of brain tumors with a priority of deploying the model in technologically underdeveloped regions. The research presents a novel architecture for precise brain tumor classification fusing pretrained ResNet152V2 and modified VGG16 models. The proposed architecture undergoes a diligent fine-tuning process that ensures fine gradients are preserved in deep neural networks, which are essential for effective brain tumor classification. The proposed solution incorporates various image processing techniques to improve image quality and achieves an astounding accuracy of 98.36% and 98.04% in Figshare and Kaggle datasets respectively. This architecture stands out for having a streamlined profile, with only 2.8 million trainable parameters. We have leveraged 8-bit quantization to produce a model of size 73.881 MB, significantly reducing it from the previous size of 289.45 MB, ensuring smooth deployment in edge devices even in resource-constrained areas. Additionally, the use of Grad-CAM improves the interpretability of the model, offering insightful information regarding its decision-making process. Owing to its high discriminative ability, this model can be a reliable option for accurate brain tumor classification.
This paper considers a multi-user multiple-input multiple-output (MU-MIMO) system where the precoding matrix is selected in a baseband unit (BBU) and then sent over a digital fronthaul to the transmitting antenna array. The fronthaul has a limited bit resolution with a known quantization behavior. We formulate a new sum rate maximization problem where the precoding matrix elements must comply with the quantizer. We solve this non-convex mixed-integer problem to local optimality by a novel iterative algorithm inspired by the classical weighted minimum mean square error (WMMSE) approach. The precoding optimization subproblem becomes an integer least-squares problem, which we solve with a new algorithm using a sphere decoding (SD) approach. We show numerically that the proposed precoding technique vastly outperforms the baseline of optimizing an infinite-resolution precoder and then quantizing it. We also develop a heuristic quantization-aware precoding that outperforms the baseline while having comparable complexity.
In this paper, we implement and evaluate a two-stage retrieval pipeline for a course recommender system that ranks courses for skill-occupation pairs. The in-production recommender system BrightFit provides course recommendations from multiple sources. Some of the course descriptions are long and noisy, while retrieval and ranking in an online system have to be highly efficient. We developed a two-step retrieval pipeline with RankT5 finetuned on MSMARCO as re-ranker. We compare two summarizers for course descriptions: a LongT5 model that we finetuned for the task, and a generative LLM (Vicuna) with in-context learning. We experiment with quantization to reduce the size of the ranking model and increase inference speed. We evaluate our rankers on two newly labelled datasets, with an A/B test, and with a user questionnaire. On the two labelled datasets, our proposed two-stage ranking with automatic summarization achieves a substantial improvement over the in-production (BM25) ranker: nDCG@10 scores improve from 0.482 to 0.684 and from 0.447 to 0.844 on the two datasets. We also achieve a 40% speed-up by using a quantized version of RankT5. The improved quality of the ranking was confirmed by the questionnaire completed by 29 respondents, but not by the A/B test. In the A/B test, a higher clickthrough rate was observed for the BM25-ranking than for the proposed two-stage retrieval. We conclude that T5-based re-ranking and summarization for online course recommendation can obtain much better effectiveness than single-step lexical retrieval, and that quantization has a large effect on RankT5. In the online evaluation, however, other factors than relevance play a role (such as speed and interpretability of the retrieval results), as well as individual preferences.
Quantizing the activations of large language models (LLMs) has been a significant challenge due to the presence of structured outliers. Most existing methods focus on the per-token or per-tensor quantization of activations, making it difficult to achieve both accuracy and hardware efficiency. To address this problem, we propose OutlierTune, an efficient per-channel post-training quantization (PTQ) method for the activations of LLMs. OutlierTune consists of two components: pre-execution of dequantization and symmetrization. The pre-execution of dequantization updates the model weights by the activation scaling factors, avoiding the internal scaling and costly additional computational overheads brought by the per-channel activation quantization. The symmetrization further reduces the quantization differences arising from the weight updates by ensuring the balanced numerical ranges across different activation channels. OutlierTune is easy to implement and hardware-efficient, introducing almost no additional computational overheads during the inference. Extensive experiments show that the proposed framework outperforms existing methods across multiple different tasks. Demonstrating better generalization, this framework improves the Int6 quantization of the instruction-tuning LLMs, such as OPT-IML, to the same level as half-precision (FP16). Moreover, we have shown that the proposed framework is 1.48x faster than the FP16 implementation while reducing approximately 2x memory usage.
3D Gaussian Splatting (3DGS) has become an emerging technique with remarkable potential in 3D representation and image rendering. However, the substantial storage overhead of 3DGS significantly impedes its practical applications. In this work, we formulate the compact 3D Gaussian learning as an end-to-end Rate-Distortion Optimization (RDO) problem and propose RDO-Gaussian that can achieve flexible and continuous rate control. RDO-Gaussian addresses two main issues that exist in current schemes: 1) Different from prior endeavors that minimize the rate under the fixed distortion, we introduce dynamic pruning and entropy-constrained vector quantization (ECVQ) that optimize the rate and distortion at the same time. 2) Previous works treat the colors of each Gaussian equally, while we model the colors of different regions and materials with learnable numbers of parameters. We verify our method on both real and synthetic scenes, showcasing that RDO-Gaussian greatly reduces the size of 3D Gaussian over 40x, and surpasses existing methods in rate-distortion performance.
The outstanding performance of large foundational models across diverse tasks-from computer vision to speech and natural language processing-has significantly increased their demand. However, storing and transmitting these models pose significant challenges due to their massive size (e.g., 350GB for GPT-3). Recent literature has focused on compressing the original weights or reducing the number of parameters required for fine-tuning these models. These compression methods typically involve constraining the parameter space, for example, through low-rank reparametrization (e.g., LoRA) or quantization (e.g., QLoRA) during model training. In this paper, we present MCNC as a novel model compression method that constrains the parameter space to low-dimensional pre-defined and frozen nonlinear manifolds, which effectively cover this space. Given the prevalence of good solutions in over-parameterized deep neural networks, we show that by constraining the parameter space to our proposed manifold, we can identify high-quality solutions while achieving unprecedented compression rates across a wide variety of tasks. Through extensive experiments in computer vision and natural language processing tasks, we demonstrate that our method, MCNC, significantly outperforms state-of-the-art baselines in terms of compression, accuracy, and/or model reconstruction time.
Communication-constrained algorithms for decentralized learning and optimization rely on local updates coupled with the exchange of compressed signals. In this context, differential quantization is an effective technique to mitigate the negative impact of compression by leveraging correlations between successive iterates. In addition, the use of error feedback, which consists of incorporating the compression error into subsequent steps, is a powerful mechanism to compensate for the bias caused by the compression. Under error feedback, performance guarantees in the literature have so far focused on algorithms employing a fusion center or a special class of contractive compressors that cannot be implemented with a finite number of bits. In this work, we propose a new decentralized communication-efficient learning approach that blends differential quantization with error feedback. The approach is specifically tailored for decentralized learning problems where agents have individual risk functions to minimize subject to subspace constraints that require the minimizers across the network to lie in low-dimensional subspaces. This constrained formulation includes consensus or single-task optimization as special cases, and allows for more general task relatedness models such as multitask smoothness and coupled optimization. We show that, under some general conditions on the compression noise, and for sufficiently small step-sizes $\mu$, the resulting communication-efficient strategy is stable both in terms of mean-square error and average bit rate: by reducing $\mu$, it is possible to keep the estimation errors small (on the order of $\mu$) without increasing indefinitely the bit rate as $\mu\rightarrow 0$. The results establish that, in the small step-size regime and with a finite number of bits, it is possible to attain the performance achievable in the absence of compression.
Federated learning (FL) is a powerful machine learning paradigm which leverages the data as well as the computational resources of clients, while protecting clients' data privacy. However, the substantial model size and frequent aggregation between the server and clients result in significant communication overhead, making it challenging to deploy FL in resource-limited wireless networks. In this work, we aim to mitigate the communication overhead by using quantization. Previous research on quantization has primarily focused on the uplink communication, employing either fixed-bit quantization or adaptive quantization methods. In this work, we introduce a holistic approach by joint uplink and downlink adaptive quantization to reduce the communication overhead. In particular, we optimize the learning convergence by determining the optimal uplink and downlink quantization bit-length, with a communication energy constraint. Theoretical analysis shows that the optimal quantization levels depend on the range of model gradients or weights. Based on this insight, we propose a decreasing-trend quantization for the uplink and an increasing-trend quantization for the downlink, which aligns with the change of the model parameters during the training process. Experimental results show that, the proposed joint uplink and downlink adaptive quantization strategy can save up to 66.7% energy compared with the existing schemes.
Vision Transformers (ViTs) have achieved remarkable performance in various image classification tasks by leveraging the attention mechanism to process image patches as tokens. However, the high computational and memory demands of ViTs pose significant challenges for deployment in resource-constrained environments. This paper introduces ViT-1.58b, a novel 1.58-bit quantized ViT model designed to drastically reduce memory and computational overhead while preserving competitive performance. ViT-1.58b employs ternary quantization, which refines the balance between efficiency and accuracy by constraining weights to {-1, 0, 1} and quantizing activations to 8-bit precision. Our approach ensures efficient scaling in terms of both memory and computation. Experiments on CIFAR-10 and ImageNet-1k demonstrate that ViT-1.58b maintains comparable accuracy to full-precision Vit, with significant reductions in memory usage and computational costs. This paper highlights the potential of extreme quantization techniques in developing sustainable AI solutions and contributes to the broader discourse on efficient model deployment in practical applications. Our code and weights are available at https://github.com/DLYuanGod/ViT-1.58b.
3D Gaussian splatting provides excellent visual quality for novel view synthesis, with fast training and real-time rendering; unfortunately, the memory requirements of this method for storing and transmission are unreasonably high. We first analyze the reasons for this, identifying three main areas where storage can be reduced: the number of 3D Gaussian primitives used to represent a scene, the number of coefficients for the spherical harmonics used to represent directional radiance, and the precision required to store Gaussian primitive attributes. We present a solution to each of these issues. First, we propose an efficient, resolution-aware primitive pruning approach, reducing the primitive count by half. Second, we introduce an adaptive adjustment method to choose the number of coefficients used to represent directional radiance for each Gaussian primitive, and finally a codebook-based quantization method, together with a half-float representation for further memory reduction. Taken together, these three components result in a 27 reduction in overall size on disk on the standard datasets we tested, along with a 1.7 speedup in rendering speed. We demonstrate our method on standard datasets and show how our solution results in significantly reduced download times when using the method on a mobile device.
Skin cancer is a major concern to public health, accounting for one-third of the reported cancers. If not detected early, the cancer has the potential for severe consequences. Recognizing the critical need for effective skin cancer classification, we address the limitations of existing models, which are often too large to deploy in areas with limited computational resources. In response, we present a knowledge distillation based approach for creating a lightweight yet high-performing classifier. The proposed solution involves fusing three models, namely ResNet152V2, ConvNeXtBase, and ViT Base, to create an effective teacher model. The teacher model is then employed to guide a lightweight student model of size 2.03 MB. This student model is further compressed to 469.77 KB using 16-bit quantization, enabling smooth incorporation into edge devices. With six-stage image preprocessing, data augmentation, and a rigorous ablation study, the model achieves an impressive accuracy of 98.75% on the HAM10000 dataset and 98.94% on the Kaggle dataset in classifying benign and malignant skin cancers. With its high accuracy and compact size, our model appears to be a potential choice for accurate skin cancer classification, particularly in resource-constrained settings.
Large language models (LLMs) have recently demonstrated remarkable performance across diverse language tasks. But their deployment is often constrained by their substantial computational and storage requirements. Quantization has emerged as a key technique for addressing this challenge, enabling the compression of large models with minimal impact on performance. The recent GPTQ algorithm, a post-training quantization (PTQ) method, has proven highly effective for compressing LLMs, sparking a wave of research that leverages GPTQ as a core component. Recognizing the pivotal role of GPTQ in the PTQ landscape, we introduce CDQuant, a simple and scalable alternative to GPTQ with improved performance. CDQuant uses greedy coordinate descent to minimize the layer-wise reconstruction loss to achieve high-quality quantized weights. Our algorithm is easy to implement and scales efficiently to models with hundreds of billions of parameters. We perform extensive evaluation on Gemma, and PaLM2 model families, and demonstrate that CDQuant consistently outperforms GPTQ in 2-4 bit weight quantization. Moreover, CDQuant improves the performance of state-of-the-art PTQ techniques such as QuIP and FrameQuant when used as a replacement for their GPTQ component, resulting in further gains in quality.
We present a simple meta quantization approach that quantizes different layers of a large language model (LLM) at different bit levels, and is independent of the underlying quantization technique. Specifically, we quantize the most important layers to higher bit precision and less important layers to lower bits. We propose two effective strategies to measure the importance of layers within LLMs: the first measures the importance of a layer based on how different its output embeddings are from the input embeddings (higher is better); the second estimates the importance of a layer using the number of layer weights that are much larger than average (smaller is better). We show that quantizing different layers at varying bits according to our importance scores results in minimal performance drop with a far more compressed model size. Finally, we present several practical key takeaways from our variable layer-wise quantization experiments: (a) LLM performance under variable quantization remains close to the original model until 25-50% of layers are moved in lower quantization using our proposed ordering but only until 5-10% if moved using no specific ordering; (b) Adding layer importance to inherently dynamic quantization techniques can further improve their performance, showing that our approach is complementary to other dynamic quantization methods; (c) Quantizing LLMs to lower bits performs substantially better than pruning unless extreme quantization (2-bit) is used; and (d) Layer-wise quantization to lower bits works better in the case of larger LLMs with more layers compared to smaller LLMs with fewer layers. Our code is publicly available at https://github.com/RazvanDu/LayerwiseQuant/.
Recent advancements in diffusion models, particularly the architectural transformation from UNet-based models to Diffusion Transformers (DiTs), significantly improve the quality and scalability of image and video generation. However, despite their impressive capabilities, the substantial computational costs of these large-scale models pose significant challenges for real-world deployment. Post-Training Quantization (PTQ) emerges as a promising solution, enabling model compression and accelerated inference for pretrained models, without the costly retraining. However, research on DiT quantization remains sparse, and existing PTQ frameworks, primarily designed for traditional diffusion models, tend to suffer from biased quantization, leading to notable performance degradation. In this work, we identify that DiTs typically exhibit significant spatial variance in both weights and activations, along with temporal variance in activations. To address these issues, we propose Q-DiT, a novel approach that seamlessly integrates two key techniques: automatic quantization granularity allocation to handle the significant variance of weights and activations across input channels, and sample-wise dynamic activation quantization to adaptively capture activation changes across both timesteps and samples. Extensive experiments conducted on ImageNet and VBench demonstrate the effectiveness of the proposed Q-DiT. Specifically, when quantizing DiT-XL/2 to W6A8 on ImageNet ($256 \times 256$), Q-DiT achieves a remarkable reduction in FID by 1.09 compared to the baseline. Under the more challenging W4A8 setting, it maintains high fidelity in image and video generation, establishing a new benchmark for efficient, high-quality quantization in DiTs. Code is available at \href{https://github.com/Juanerx/Q-DiT}{https://github.com/Juanerx/Q-DiT}.
Recent expansions in multimedia devices gather enormous amounts of real-time images for processing and inference. The images are first compressed using compression schemes, like JPEG, to reduce storage costs and power for transmitting the captured data. Due to inherent error resilience and imperceptibility in images, JPEG can be approximated to reduce the required computation power and area. This work demonstrates the first end-to-end approximation computing-based optimization of JPEG hardware using i) an approximate division realized using bit-shift operators to reduce the complexity of the quantization block, ii) loop perforation, and iii) precision scaling on top of a multiplier-less fast DCT architecture to achieve an extremely energy-efficient JPEG compression unit which will be a perfect fit for power/bandwidth-limited scenario. Furthermore, a gradient descent-based heuristic composed of two conventional approximation strategies, i.e., Precision Scaling and Loop Perforation, is implemented for tuning the degree of approximation to trade off energy consumption with the quality degradation of the decoded image. The entire RTL design is coded in Verilog HDL, synthesized, mapped to TSMC 65nm CMOS technology, and simulated using Cadence Spectre Simulator under 25$^{\circ}$\textbf{C}, TT corner. The approximate division approach achieved around $\textbf{28\%}$ reduction in the active design area. The heuristic-based approximation technique combined with accelerator optimization achieves a significant energy reduction of $\textbf{36\%}$ for a minimal image quality degradation of $\textbf{2\%}$ SAD. Simulation results also show that the proposed architecture consumes 15uW at the DCT and quantization stages to compress a colored 480p image at 6fps.
To maximize the received power at a user equipment, the problem of optimizing a reconfigurable intelligent surface (RIS) with a limited phase range R < 2{\pi} and nonuniform discrete phase shifts with adjustable gains is addressed. Necessary and sufficient conditions to achieve this maximization are given. These conditions are employed in two algorithms to achieve the global optimum in linear time for R {\ge} {\pi} and R < {\pi}, where R is the limited RIS phase range. With a total number of N(2K + 1) complex vector additions, it is shown for R {\ge} {\pi} and R < {\pi} that the global optimality is achieved in NK or fewer and N(K + 1) or fewer steps, respectively, where N is the number of RIS elements and K is the number of discrete phase shifts which may be placed nonuniformly over the limited phase range R. In addition, we define two quantization algorithms that we call nonuniform polar quantization (NPQ) algorithm and extended nonuniform polar quantization (ENPQ) algorithm, where the latter is a novel quantization algorithm for RISs with a significant phase range restriction, i.e., R < {\pi}. With NPQ, we provide a closed-form solution for the approximation ratio with which an arbitrary set of nonuniform discrete phase shifts can approximate the continuous solution. We also show that with a phase range limitation, equal separation among the nonuniform discrete phase shifts maximizes the normalized performance. Furthermore, we show that the gain of using K {\ge} 3 with R < {\pi}/2 and K {\ge} 4 with R < {\pi} is only marginal. Finally, we prove that when R < 2{\pi}/3, ON/OFF selection for the RIS elements brings significant performance compared to the case when the RIS elements are strictly ON.
Efficient adaption of large language models (LLMs) on edge devices is essential for applications requiring continuous and privacy-preserving adaptation and inference. However, existing tuning techniques fall short because of the high computation and memory overheads. To this end, we introduce a computation- and memory-efficient LLM tuning framework, called Edge-LLM, to facilitate affordable and effective LLM adaptation on edge devices. Specifically, Edge-LLM features three core components: (1) a layer-wise unified compression (LUC) technique to reduce the computation overhead by generating layer-wise pruning sparsity and quantization bit-width policies, (2) an adaptive layer tuning and voting scheme to reduce the memory overhead by reducing the backpropagation depth, and (3) a complementary hardware scheduling strategy to handle the irregular computation patterns introduced by LUC and adaptive layer tuning, thereby achieving efficient computation and data movements. Extensive experiments demonstrate that Edge-LLM achieves a 2.92x speed up and a 4x memory overhead reduction as compared to vanilla tuning methods with comparable task accuracy. Our code is available at https://github.com/GATECH-EIC/Edge-LLM
The high power consumption and latency-sensitive deployments of large language models (LLMs) have motivated efficiency techniques like quantization and sparsity. Contextual sparsity, where the sparsity pattern is input-dependent, is crucial in LLMs because the permanent removal of attention heads or neurons from LLMs can significantly degrade accuracy. Prior work has attempted to model contextual sparsity using neural networks trained to predict activation magnitudes, which can be used to dynamically prune structures with low predicted activation magnitude. In this paper, we look beyond magnitude-based pruning criteria to assess attention head and neuron importance in LLMs. We develop a novel predictor called ShadowLLM, which can shadow the LLM behavior and enforce better sparsity patterns, resulting in over 15% improvement in end-to-end accuracy compared to prior methods. In addition, ShadowLLM achieves up to a 20% speed-up over the state-of-the-art DejaVu framework. These enhancements are validated on Llama-2 and OPT models with up to 30 billion parameters. Our code is available at \href{https://github.com/abdelfattah-lab/shadow_llm/}{ShadowLLM}.
While large language models (LLMs) excel in many domains, their complexity and scale challenge deployment in resource-limited environments. Current compression techniques, such as parameter pruning, often fail to effectively utilize the knowledge from pruned parameters. To address these challenges, we propose Manifold-Based Knowledge Alignment and Layer Merging Compression (MKA), a novel approach that uses manifold learning and the Normalized Pairwise Information Bottleneck (NPIB) measure to merge similar layers, reducing model size while preserving essential performance. We evaluate MKA on multiple benchmark datasets and various LLMs. Our findings show that MKA not only preserves model performance but also achieves substantial compression ratios, outperforming traditional pruning methods. Moreover, when coupled with quantization, MKA delivers even greater compression. Specifically, on the MMLU dataset using the Llama3-8B model, MKA achieves a compression ratio of 43.75% with a minimal performance decrease of only 2.82\%. The proposed MKA method offers a resource-efficient and performance-preserving model compression technique for LLMs.
Emergent Large Language Models (LLMs) use their extraordinary performance and powerful deduction capacity to discern from traditional language models. However, the expenses of computational resources and storage for these LLMs are stunning, quantization then arises as a trending conversation. To address accuracy decay caused by quantization, two streams of works in post-training quantization methods stand out. One uses other weights to compensate existing quantization error, while the other transfers the quantization difficulty to other parts in the model. Combining both merits, we introduce Learnable Singular value Increment (LSI) as an advanced solution. LSI uses Singular Value Decomposition to extract singular values of the weights and make them learnable to help weights compensate each other conditioned on activation. Incorporating LSI with existing techniques, we achieve state-of-the-art performance in diverse quantization settings, no matter in weight-only, weight-activation or extremely low bit scenarios. By unleashing the potential of LSI, efficient finetuning on quantized model is no longer a prohibitive problem.