-
In-BoXBART: Get Instructions into Biomedical Multi-Task Learning
Authors:
Mihir Parmar,
Swaroop Mishra,
Mirali Purohit,
Man Luo,
M. Hassan Murad,
Chitta Baral
Abstract:
Single-task models have proven pivotal in solving specific tasks; however, they have limitations in real-world applications where multi-tasking is necessary and domain shifts are exhibited. Recently, instructional prompts have shown significant improvement towards multi-task generalization; however, the effect of instructional prompts and Multi-Task Learning (MTL) has not been systematically studi…
▽ More
Single-task models have proven pivotal in solving specific tasks; however, they have limitations in real-world applications where multi-tasking is necessary and domain shifts are exhibited. Recently, instructional prompts have shown significant improvement towards multi-task generalization; however, the effect of instructional prompts and Multi-Task Learning (MTL) has not been systematically studied in the biomedical domain. Motivated by this, this paper explores the impact of instructional prompts for biomedical MTL. We introduce the BoX, a collection of 32 instruction tasks for Biomedical NLP across (X) various categories. Using this meta-dataset, we propose a unified model termed In-BoXBART, that can jointly learn all tasks of the BoX without any task-specific modules. To the best of our knowledge, this is the first attempt to propose a unified model in the biomedical domain and use instructions to achieve generalization across several biomedical tasks. Experimental results indicate that the proposed model: 1) outperforms the single-task baseline by ~3% and multi-task (without instruction) baseline by ~18% on an average, and 2) shows ~23% improvement compared to the single-task baseline in few-shot learning (i.e., 32 instances per task) on an average. Our analysis indicates that there is significant room for improvement across tasks in the BoX, implying the scope for future research direction.
△ Less
Submitted 15 April, 2022;
originally announced April 2022.
-
Power Consumption and Energy-Efficiency for In-Band Full-Duplex Wireless Systems
Authors:
Murad Murad,
Ahmed M. Eltawil
Abstract:
This paper presents an analytical model of power consumption for In-Band Full-Duplex (IBFD) Wireless Local-Area Networks (WLANs). Energy-efficiency is compared for both Half-Duplex (HD) and IBFD networks. The presented analytical model closely matches the results generated by simulation. For a given traffic scenario, IBFD systems exhibit higher power consumption, however at improved energy efficie…
▽ More
This paper presents an analytical model of power consumption for In-Band Full-Duplex (IBFD) Wireless Local-Area Networks (WLANs). Energy-efficiency is compared for both Half-Duplex (HD) and IBFD networks. The presented analytical model closely matches the results generated by simulation. For a given traffic scenario, IBFD systems exhibit higher power consumption, however at improved energy efficiency when compared to equivalent HD WLANs.
△ Less
Submitted 23 April, 2019;
originally announced April 2019.
-
Performance Analysis and Enhancements for In-Band Full-Duplex Wireless Local Area Networks
Authors:
Murad Murad,
Ahmed M. Eltawil
Abstract:
In-Band Full-Duplex (IBFD) is a technique that enables a wireless node to simultaneously transmit a signal and receive another on the same assigned frequency. Thus, IBFD wireless systems can provide up to twice the channel capacity compared to conventional Half-Duplex (HD) systems. In order to study the feasibility of IBFD networks, reliable models are needed to capture anticipated benefits of IBF…
▽ More
In-Band Full-Duplex (IBFD) is a technique that enables a wireless node to simultaneously transmit a signal and receive another on the same assigned frequency. Thus, IBFD wireless systems can provide up to twice the channel capacity compared to conventional Half-Duplex (HD) systems. In order to study the feasibility of IBFD networks, reliable models are needed to capture anticipated benefits of IBFD above the physical layer (PHY). In this paper, an accurate analytical model based on Discrete-Time Markov Chain (DTMC) analysis for IEEE 802.11 Distributed Coordination Function (DCF) with IBFD capabilities is proposed. The model captures all parameters necessary to calculate important performance metrics which quantify enhancements introduced as a result of IBFD solutions. Additionally, two frame aggregation schemes for Wireless Local Area Networks (WLANs) with IBFD features are proposed to increase the efficiency of data transmission. Matching analytical and simulation results with less than 1% average errors confirm that the proposed frame aggregation schemes further improve the overall throughput by up to 24% and reduce latency by up to 47% in practical IBFD-WLANs. More importantly, the results assert that IBFD transmission can only reduce latency to a suboptimal point in WLANs, but frame aggregation is necessary to minimize it.
△ Less
Submitted 27 March, 2019;
originally announced March 2019.
-
Where is My Next Hop ? The Case of Indian Ocean Islands
Authors:
Xavier Nicolay,
Rehan Noordally,
Pascal Anelli,
Nour Mohammad Murad,
Tahiry Razafindralambo
Abstract:
Internet has become a foundation of our modern society. However, all regions or countries do not have the same Internet access regarding quality especially in the Indian Ocean Area (IOA). To improve this quality it is important to have a deep knowledge of the Internet physical and logical topology and associated performance. However, these knowledges are not shared by Internet service providers. I…
▽ More
Internet has become a foundation of our modern society. However, all regions or countries do not have the same Internet access regarding quality especially in the Indian Ocean Area (IOA). To improve this quality it is important to have a deep knowledge of the Internet physical and logical topology and associated performance. However, these knowledges are not shared by Internet service providers. In this paper, we describe a large scale measurement study in which we deploy probes in different IOA countries, we generate network traces, develop a tool to extract useful information and analyze these information. We show that most of the IOA traffic exits through one point even if there exists multiple exit points.
△ Less
Submitted 28 September, 2017; v1 submitted 21 July, 2017;
originally announced July 2017.
