-
Signed Binarization: Unlocking Efficiency Through Repetition-Sparsity Trade-Off
Authors:
Sachit Kuhar,
Yash Jain,
Alexey Tumanov
Abstract:
Efficient inference of Deep Neural Networks (DNNs) on resource-constrained edge devices is essential. Quantization and sparsity are key algorithmic techniques that translate to repetition and sparsity within tensors at the hardware-software interface. This paper introduces the concept of repetition-sparsity trade-off that helps explain computational efficiency during inference. We propose Signed B…
▽ More
Efficient inference of Deep Neural Networks (DNNs) on resource-constrained edge devices is essential. Quantization and sparsity are key algorithmic techniques that translate to repetition and sparsity within tensors at the hardware-software interface. This paper introduces the concept of repetition-sparsity trade-off that helps explain computational efficiency during inference. We propose Signed Binarization, a unified co-design framework that synergistically integrates hardware-software systems, quantization functions, and representation learning techniques to address this trade-off. Our results demonstrate that Signed Binarization is more accurate than binarization with the same number of non-zero weights. Detailed analysis indicates that signed binarization generates a smaller distribution of effectual (non-zero) parameters nested within a larger distribution of total parameters, both of the same type, for a DNN block. Finally, our approach achieves a 26% speedup on real hardware, doubles energy efficiency, and reduces density by 2.8x compared to binary methods for ResNet 18, presenting an alternative solution for deploying efficient models in resource-limited environments.
△ Less
Submitted 3 December, 2023;
originally announced December 2023.
-
Learning to Discern: Imitating Heterogeneous Human Demonstrations with Preference and Representation Learning
Authors:
Sachit Kuhar,
Shuo Cheng,
Shivang Chopra,
Matthew Bronars,
Danfei Xu
Abstract:
Practical Imitation Learning (IL) systems rely on large human demonstration datasets for successful policy learning. However, challenges lie in maintaining the quality of collected data and addressing the suboptimal nature of some demonstrations, which can compromise the overall dataset quality and hence the learning outcome. Furthermore, the intrinsic heterogeneity in human behavior can produce e…
▽ More
Practical Imitation Learning (IL) systems rely on large human demonstration datasets for successful policy learning. However, challenges lie in maintaining the quality of collected data and addressing the suboptimal nature of some demonstrations, which can compromise the overall dataset quality and hence the learning outcome. Furthermore, the intrinsic heterogeneity in human behavior can produce equally successful but disparate demonstrations, further exacerbating the challenge of discerning demonstration quality. To address these challenges, this paper introduces Learning to Discern (L2D), an offline imitation learning framework for learning from demonstrations with diverse quality and style. Given a small batch of demonstrations with sparse quality labels, we learn a latent representation for temporally embedded trajectory segments. Preference learning in this latent space trains a quality evaluator that generalizes to new demonstrators exhibiting different styles. Empirically, we show that L2D can effectively assess and learn from varying demonstrations, thereby leading to improved policy performance across a range of tasks in both simulations and on a physical robot.
△ Less
Submitted 22 October, 2023;
originally announced October 2023.
-
Signed Binary Weight Networks
Authors:
Sachit Kuhar,
Alexey Tumanov,
Judy Hoffman
Abstract:
Efficient inference of Deep Neural Networks (DNNs) is essential to making AI ubiquitous. Two important algorithmic techniques have shown promise for enabling efficient inference - sparsity and binarization. These techniques translate into weight sparsity and weight repetition at the hardware-software level enabling the deployment of DNNs with critically low power and latency requirements. We propo…
▽ More
Efficient inference of Deep Neural Networks (DNNs) is essential to making AI ubiquitous. Two important algorithmic techniques have shown promise for enabling efficient inference - sparsity and binarization. These techniques translate into weight sparsity and weight repetition at the hardware-software level enabling the deployment of DNNs with critically low power and latency requirements. We propose a new method called signed-binary networks to improve efficiency further (by exploiting both weight sparsity and weight repetition together) while maintaining similar accuracy. Our method achieves comparable accuracy on ImageNet and CIFAR10 datasets with binary and can lead to 69% sparsity. We observe real speedup when deploying these models on general-purpose devices and show that this high percentage of unstructured sparsity can lead to a further reduction in energy consumption on ASICs.
△ Less
Submitted 4 December, 2023; v1 submitted 24 November, 2022;
originally announced November 2022.
-
Effect of Antral Motility on Food Hydrolysis and Gastric Emptying from the Stomach: Insights from Computational Models
Authors:
Sharun Kuhar,
Jae Ho Lee,
Jung-Hee Seo,
Pankaj J Pasricha,
Rajat Mittal
Abstract:
The peristaltic motion of the stomach walls combines with the secretion of enzymes to initiate the process that breaks down food. Computational modelling of this phenomenon can help reveal the details that would be hard to capture via in-vivo or in-vitro means. In this study, the digestion of a liquid meal containing protein is simulated in a human-stomach model based on imaging data. Pepsin, the…
▽ More
The peristaltic motion of the stomach walls combines with the secretion of enzymes to initiate the process that breaks down food. Computational modelling of this phenomenon can help reveal the details that would be hard to capture via in-vivo or in-vitro means. In this study, the digestion of a liquid meal containing protein is simulated in a human-stomach model based on imaging data. Pepsin, the gastric enzyme for protein hydrolysis, is secreted from the proximal region of the stomach walls and allowed to react with the contents of the stomach. The jet velocities, the emptying rate, and the extent of hydrolysis are quantified for a control case, and also for three other cases of reduced motility with varying peristaltic amplitudes. The findings quantify the effect of motility on the rate of food breakdown and emptying, and correlate the observations with the mixing in the stomach induced by the antral contraction waves.
△ Less
Submitted 13 August, 2022;
originally announced August 2022.
-
Offline Visual Representation Learning for Embodied Navigation
Authors:
Karmesh Yadav,
Ram Ramrakhya,
Arjun Majumdar,
Vincent-Pierre Berges,
Sachit Kuhar,
Dhruv Batra,
Alexei Baevski,
Oleksandr Maksymets
Abstract:
How should we learn visual representations for embodied agents that must see and move? The status quo is tabula rasa in vivo, i.e. learning visual representations from scratch while also learning to move, potentially augmented with auxiliary tasks (e.g. predicting the action taken between two successive observations). In this paper, we show that an alternative 2-stage strategy is far more effectiv…
▽ More
How should we learn visual representations for embodied agents that must see and move? The status quo is tabula rasa in vivo, i.e. learning visual representations from scratch while also learning to move, potentially augmented with auxiliary tasks (e.g. predicting the action taken between two successive observations). In this paper, we show that an alternative 2-stage strategy is far more effective: (1) offline pretraining of visual representations with self-supervised learning (SSL) using large-scale pre-rendered images of indoor environments (Omnidata), and (2) online finetuning of visuomotor representations on specific tasks with image augmentations under long learning schedules. We call this method Offline Visual Representation Learning (OVRL). We conduct large-scale experiments - on 3 different 3D datasets (Gibson, HM3D, MP3D), 2 tasks (ImageNav, ObjectNav), and 2 policy learning algorithms (RL, IL) - and find that the OVRL representations lead to significant across-the-board improvements in state of art, on ImageNav from 29.2% to 54.2% (+25% absolute, 86% relative) and on ObjectNav from 18.1% to 23.2% (+5.1% absolute, 28% relative). Importantly, both results were achieved by the same visual encoder generalizing to datasets that were not seen during pretraining. While the benefits of pretraining sometimes diminish (or entirely disappear) with long finetuning schedules, we find that OVRL's performance gains continue to increase (not decrease) as the agent is trained for 2 billion frames of experience.
△ Less
Submitted 27 April, 2022;
originally announced April 2022.
-
Computational model of drug dissolution in the stomach: effects of posture and gastroparesis on drug bioavailability
Authors:
Jae H. Lee,
Sharun Kuhar,
Jung-Hee Seo,
Pankaj J. Pasricha,
Rajat Mittal
Abstract:
The oral route is the most common choice for drug administration because of convenience, low cost, and high patient compliance, but is also a complex route. The rate of dissolution and gastric emptying of the dissolved active pharmaceutical ingredient (API) into the duodenum is modulated by factors such as gastric motility, but current in-vitro procedures for assessing drug dissolution are limited…
▽ More
The oral route is the most common choice for drug administration because of convenience, low cost, and high patient compliance, but is also a complex route. The rate of dissolution and gastric emptying of the dissolved active pharmaceutical ingredient (API) into the duodenum is modulated by factors such as gastric motility, but current in-vitro procedures for assessing drug dissolution are limited in their ability to recapitulate this process. This is particularly relevant for disease conditions, such as gastroparesis, that alter the anatomy and/or physiology of the stomach. In this study we employ a biomimetic in-silico simulator based on the realistic anatomy and morphology of the stomach, to investigate the effect of body posture and stomach motility on drug bioavailability. The simulations show that changes in posture can have a significant (up to 83%) effect on the emptying rate of the API into the duodenum. Similarly, reduction in antral contractility associated with gastroparesis can also significantly reduce the dissolution of the pill as well as emptying of the API into the duodenum. The simulations show that for an equivalent motility index, reduction in gastric emptying due to neuropathic gastroparesis is larger by a factor of about five compared to myopathic gastroparesis.
△ Less
Submitted 21 January, 2022;
originally announced January 2022.
