Model compression methods are used to reduce the computation and energy requirements for Large Language Models (LLMs). Quantization Aware Training (QAT), an effective model compression method, is proposed to reduce performance degradation after quantization. To further minimize this degradation, we introduce two continuous approximations to the QAT process on the rounding function, traditionally approximated by the Straight-Through Estimator (STE), and the clamping function. By applying both methods, the perplexity (PPL) on the WikiText-v2 dataset of the quantized model reaches 9.0815, outperforming 9.9621 by the baseline. Also, we achieve a 2.76% improvement on BoolQ, and a 5.47% improvement on MMLU, proving that the step sizes and weights can be learned more accurately with our approach. Our method achieves better performance with the same precision, model size, and training setup, contributing to the development of more energy-efficient LLMs technology that aligns with global sustainability goals.

Early wildfire detection in remote and forest areas is crucial for minimizing devastation and preserving ecosystems. Autonomous drones offer agile access to remote, challenging terrains, equipped with advanced imaging technology that delivers both high-temporal and detailed spatial resolution, making them valuable assets in the early detection and monitoring of wildfires. However, the limited computation and battery resources of Unmanned Aerial Vehicles (UAVs) pose significant challenges in implementing robust and efficient image classification models. Current works in this domain often operate offline, emphasizing the need for solutions that can perform inference in real time, given the constraints of UAVs. To address these challenges, this paper aims to develop a real-time image classification and fire segmentation model. It presents a comprehensive investigation into hardware acceleration using the Jetson Nano P3450 and the implications of TensorRT, NVIDIA's high-performance deep-learning inference library, on fire classification accuracy and speed. The study includes implementations of Quantization Aware Training (QAT), Automatic Mixed Precision (AMP), and post-training mechanisms, comparing them against the latest baselines for fire segmentation and classification. All experiments utilize the FLAME dataset - an image dataset collected by low-altitude drones during a prescribed forest fire. This work contributes to the ongoing efforts to enable real-time, on-board wildfire detection capabilities for UAVs, addressing speed and the computational and energy constraints of these crucial monitoring systems. The results show a 13% increase in classification speed compared to similar models without hardware optimization. Comparatively, loss and accuracy are within 1.225% of the original values.

Optical neural architectures (ONAs) use coding elements with optimized physical parameters to perform intelligent measurements. However, fabricating ONAs while maintaining design performances is challenging. Limitations in fabrication techniques often limit the realizable precision of the trained parameters. Physical constraints may also limit the range of values the physical parameters can hold. Thus, ONAs should be trained within the implementable constraints. However, such physics-based constraints reduce the training objective to a constrained optimization problem, making it harder to optimize with existing gradient-based methods. To alleviate these critical issues that degrade performance from simulation to realization we propose a physics-informed quantization-aware training framework. Our approach accounts for the physical constraints during the training process, leading to robust designs. We evaluate our approach on an ONA proposed in the literature, named a diffractive deep neural network (D2NN), for all-optical phase imaging and for classification of phase objects. With extensive experiments on different quantization levels and datasets, we show that our approach leads to ONA designs that are robust to quantization noise.

Quantization refers to techniques for performing computations and storing tensors at lower bitwidths than floating point precision. A quantized model executes some or all of the operations on tensors with reduced precision rather than full precision (floating point) values. This allows for a more compact model representation and the use of high performance vectorized operations on many hardware platforms. PyTorch supports INT8 quantization compared to typical FP32 models allowing for a 4x reduction in the model size and a 4x reduction in memory bandwidth requirements. Hardware support for INT8 computations is typically 2 to 4 times faster compared to FP32 compute.

At present, the quantification methods of neural network models are mainly divided into post-training quantization (PTQ) and quantization aware training (QAT). Post-training quantization only need a small part of the data to complete the quantification process, but the performance of its quantitative model is not as good as the quantization aware training. This paper presents a novel quantification method called Attention Round. This method gives parameters w the opportunity to be mapped to all possible quantized values, rather than just the two quantized values nearby w in the process of quantization. The probability of being mapped to different quantified values is negatively correlated with the distance between the quantified values and w, and decay with a Gaussian function. In addition, this paper uses the lossy coding length as a measure to assign bit widths to the different layers of the model to solve the problem of mixed precision quantization, which effectively avoids to solve combinatorial optimization problem. This paper also performs quantitative experiments on different models, the results confirm the effectiveness of the proposed method. For ResNet18 and MobileNetV2, the post-training quantization proposed in this paper only require 1,024 training data and 10 minutes to complete the quantization process, which can achieve quantization performance on par with quantization aware training.

Typically, you need high computing power to deploy and run a machine learning model. GPU (Graphics Processing Unit) is designed to perform floating-point operations, unlike CPU, which fulfills more diverse tasks. GPUs help implement machine learning algorithms because of their ability to perform complex mathematical calculations. Microcontrollers do not contain enough resources to run the typical machine learning algorithms. The computing power of microcontrollers is much lower than GPUs, which is why a standard ML algorithm is not executable on such resource-constraint hardware.

Pre-trained language models like Ernie or Bert are currently used in many applications. These models come with a set of pre-trained weights typically obtained in unsupervised/self-supervised modality on a huge amount of data. After that, they are fine-tuned on a specific task. Applications then use these models for inference, and often some additional constraints apply, like low power-budget or low latency between input and output. The main avenue to meet these additional requirements for the inference settings, is to use low precision computation (e.g. INT8 rather than FP32), but this comes with a cost of deteriorating the functional performance (e.g. accuracy) of the model. Some approaches have been developed to tackle the problem and go beyond the limitations of the PTO (Post-Training Quantization), more specifically the QAT (Quantization Aware Training, see [4]) is a procedure that interferes with the training process in order to make it affected (or simply disturbed) by the quantization phase during the training itself. Besides QAT, recently Intel-Habana Labs have proposed an additional and more direct way to make the training results more robust to subsequent quantization which uses a regularizer, therefore changing the loss function that drives the training procedure. But their proposal does not work out-of-the-box for pre-trained models like Ernie, for example. In this short paper we show why this is not happening (for the Ernie case) and we propose a very basic way to deal with it, sharing as well some initial results (increase in final INT8 accuracy) that might be of interest to practitioners willing to use Ernie in their applications, in low precision regime.

Real-world applications of Deep Neural Networks are increasing by the day as we are learning to make use of Artificial Intelligence to accomplish various simple and complex tasks. However, the problem with Deep Neural Networks is that they involve too many parameters due to which they require powerful computation devices and large memory storage. This makes it almost impossible to run on devices with lower computation power such as Android and other low-power edge devices. Optimization techniques such as Quantization can be utilized to solve this problem. With the help of different quantization techniques, we can reduce the precision of our parameters from float to lower precision such as int8, resulting in efficient computation and less amount of storage.

Modern Neural Networks are eminent in achieving state of the art performance on tasks under Computer Vision, Natural Language Processing and related verticals. However, they are notorious for their voracious memory and compute appetite which further obstructs their deployment on resource limited edge devices. In order to achieve edge deployment, researchers have developed pruning and quantization algorithms to compress such networks without compromising their efficacy. Such compression algorithms are broadly experimented on standalone CNN and RNN architectures while in this work, we present an unconventional end to end compression pipeline of a CNN-LSTM based Image Captioning model. The model is trained using VGG16 or ResNet50 as an encoder and an LSTM decoder on the flickr8k dataset. We then examine the effects of different compression architectures on the model and design a compression architecture that achieves a 73.1% reduction in model size, 71.3% reduction in inference time and a 7.7% increase in BLEU score as compared to its uncompressed counterpart.

Recent work has shown that quantization can help reduce the memory, compute, and energy demands of deep neural networks without significantly harming their quality. However, whether these prior techniques, applied traditionally to image-based models, work with the same efficacy to the sequential decision making process in reinforcement learning remains an unanswered question. To address this void, we conduct the first comprehensive empirical study that quantifies the effects of quantization on various deep reinforcement learning policies with the intent to reduce their computational resource demands. We apply techniques such as post-training quantization and quantization aware training to a spectrum of reinforcement learning tasks (such as Pong, Breakout, BeamRider and more) and training algorithms (such as PPO, A2C, DDPG, and DQN). Across this spectrum of tasks and learning algorithms, we show that policies can be quantized to 6-8 bits of precision without loss of accuracy. We also show that certain tasks and reinforcement learning algorithms yield policies that are more difficult to quantize due to their effect of widening the models' distribution of weights and that quantization aware training consistently improves results over post-training quantization and oftentimes even over the full precision baseline. Finally, we demonstrate real-world applications of quantization for reinforcement learning. We use half-precision training to train a Pong model 50% faster, and we deploy a quantized reinforcement learning based navigation policy to an embedded system, achieving an 18$\times$ speedup and a 4$\times$ reduction in memory usage over an unquantized policy.