| 1. |
- Haberer, Janek, et al.
(författare)
-
Activation sparsity and dynamic pruning for split computing in edge AI
- 2022
-
Ingår i: DistributedML 2022 - Proceedings of the 3rd International Workshop on Distributed Machine Learning, Part of CoNEXT 2022. - New York, NY, USA : ACM. ; , s. 30-36
-
Konferensbidrag (refereegranskat)abstract
- Deep neural networks are getting larger and, therefore, harder to deploy on constrained IoT devices. Split computing provides a solution by splitting a network and placing the first few layers on the IoT device. The output of these layers is transmitted to the cloud where inference continues. Earlier works indicate a degree of high sparsity in intermediate activation outputs, this paper analyzes and exploits activation sparsity to reduce the network communication overhead when transmitting intermediate data to the cloud. Specifically, we analyze the intermediate activations of two early layers in ResNet-50 on CIFAR-10 and ImageNet, focusing on sparsity to guide the process of choosing a splitting point. We employ dynamic pruning of activations and feature maps and find that sparsity is very dependent on the size of a layer, and weights do not correlate with activation sparsity in convolutional layers. Additionally, we show that sparse intermediate outputs can be compressed by a factor of 3.3X at an accuracy loss of 1.1% without any fine-tuning. When adding fine-tuning, the compression factor increases up to 14X at a total accuracy loss of 1%.
|
|
| 2. |
|
|
| 3. |
- Harms, Laura, 1991, et al.
(författare)
-
Grace: Low-Cost Time-Synchronized GPIO Tracing for IoT Testbeds
- 2022
-
Ingår i: Proceedings - 18th Annual International Conference on Distributed Computing in Sensor Systems, DCOSS 2022. ; , s. 9-16
-
Konferensbidrag (refereegranskat)abstract
- Testbeds have become a vital tool for evaluating and benchmarking applications and algorithms in the Internet of Things (IoT). Testbeds commonly consist of low-power IoT devices augmented with observer nodes providing control, logging, and often also power-profiling. Today, the research community operates numerous testbeds, sometimes with hundreds of IoT nodes, to allow for detailed and large-scale evaluation. Most testbeds, however, lack opportunities for tracing distributed program execution with high accuracy in time, for example, via minimally invasive, distributed GPIO tracing. And the ones that do, like Flocklab, are built from custom hardware, which is often too complex, inflexible, or expensive to use for other research groups.This paper closes this gap and introduces Grace, a low-cost, retrofittable, distributed, and time-synchronized GPIO tracing system built from off-the-shelf components, costing less than €20 per node. Grace extends observer nodes in a testbed with (1) time-synchronization via wireless sub-GHz transceivers and (2) logic analyzers for GPIO tracing and logging, enabling time-synchronized GPIO tracing at a frequency of up to 8 MHz. We deploy Grace in a testbed and show that it achieves an average time synchronization error between nodes of 1.53 µs.
|
|
| 4. |
- Harms, Laura, 1991, et al.
(författare)
-
Grace: Low-cost time-synchronized GPIO tracing for IoT testbeds
- 2023
-
Ingår i: Computer Networks. - : Elsevier BV. - 1389-1286. ; 228
-
Tidskriftsartikel (refereegranskat)abstract
- Testbeds have become a vital tool for evaluating and benchmarking applications and algorithms in the Internet of Things (IoT). IoT testbeds commonly consist of low-power IoT devices augmented with observer nodes providing control, debugging, logging, and often also power-profiling capabilities. Today, the research community operates numerous testbeds, sometimes with hundreds of IoT nodes, to allow for detailed and large-scale evaluation. Most testbeds, however, lack opportunities for tracing distributed program execution with high accuracy in time, for example, via minimally invasive, distributed GPIO tracing. And the ones that do, like Flocklab, are built from custom hardware, which is often too complex, inflexible, or expensive to use for other research groups. This paper closes this gap and introduces Grace, a low-cost, retrofittable, distributed, and time-synchronized GPIO tracing system built from off-the-shelf components, costing less than €20 per node. Grace extends observer nodes in a testbed with (1) time-synchronization via wireless sub-GHz transceivers and (2) logic analyzers for GPIO tracing and logging, enabling time-synchronized GPIO tracing at a frequency of up to 8 MHz. We deploy Grace in a testbed and evaluate it, showing that it achieves an average time synchronization error between nodes of 1.53 μs using a single time source, and 15.3 μs between nodes using different time sources, sufficient for most IoT applications.
|
|
| 5. |
- Harms, Laura, 1991, et al.
(författare)
-
MASTER: Long-Term Stable Routing and Scheduling in Low-Power Wireless Networks
- 2020
-
Ingår i: 16TH ANNUAL INTERNATIONAL CONFERENCE ON DISTRIBUTED COMPUTING IN SENSOR SYSTEMS (DCOSS 2020). - 2325-2936. ; , s. 86-94
-
Konferensbidrag (refereegranskat)abstract
- Wireless Sensor-Actuator Networks (WSANs) are an important driver for the Industrial Internet of Things (IIoT) as they easily retrofit existing industrial infrastructure. Industrial applications require these networks to provide stable communication with high reliability and guaranteed low latency. A common way is using a central scheduler to plan transmissions and routes so that all packets are delivered before a deadline. However, existing centralized schedulers are only able to achieve high reliability in the absence of interference. This limitation lowers the feasibility of using centralized schedulers in most environments susceptible to interference. This paper addresses the challenge of stable, centrally scheduled communication in low-power wireless networks susceptible to interference. We introduce MASTER, a centralized scheduler and router, for IEEE 802.15.4 TSCH (Time-Slotted Channel Hopping). MASTER uses Sliding Windows, a novel transmission strategy, which builds on flow-based retransmissions instead of link-based ones. We show in our experimental evaluation that MASTER with Sliding Windows achieves routing and scheduling stability for over 24 hours with end-to-end reliability of over 99.6%. Moreover, we show that MASTER outperforms Orchestra, a state-of-the-art autonomous scheduler, in terms of latency by a factor of 8 while achieving similar reliability under a slight duty-cycle increase.
|
|
| 6. |
- Harms, Laura, 1991, et al.
(författare)
-
Opportunistic Routing and Synchronous Transmissions Meet TSCH
- 2021
-
Ingår i: Proceedings - Conference on Local Computer Networks, LCN. ; 2021-October, s. 107-114
-
Konferensbidrag (refereegranskat)abstract
- Low-power wireless networking commonly uses either Time-Slotted Channel Hopping (TSCH), synchronous transmissions, or opportunistic routing. All three of these different, orthogonal approaches strive for efficient and reliable communication but follow different trajectories. With this paper, we combine these concepts into one protocol: AUTOBAHN. AUTOBAHN merges TSCH scheduling with opportunistically routed, synchronous transmissions. This opens the possibility to create long-term stable schedules overcoming local interference. We prove the stability of schedules over several days in our experimental evaluation. Moreover, AUTOBAHN outperforms the autonomous scheduler Orchestra under interference in terms of reliability by 13.9 percentage points and in terms of latency by a factor of 9 under a minor duty cycle increase of 2.1 percentage points.
|
|
| 7. |
- Harms, Laura, 1991, et al.
(författare)
-
(POSTER) OVERTAKE: Opportunistic Routing and Concurrent Transmissions for TSCH
- 2020
-
Ingår i: 16TH ANNUAL INTERNATIONAL CONFERENCE ON DISTRIBUTED COMPUTING IN SENSOR SYSTEMS (DCOSS 2020). - 2325-2936. ; , s. 141-143
-
Konferensbidrag (refereegranskat)abstract
- In this paper, we present OVERTAKE, an opportunistic routing protocol for Time-Slotted Channel Hopping (TSCH). OVERTAKE combines (1) opportunistic routing, (2) concurrent transmissions and (3) TSCH. We show that this novel combination enables low-latency, central scheduling withstanding node failures. Our initial results show its ability to withstand node failures of up to 40% of nodes of a flow while keeping minimal latency.
|
|
| 8. |
- Harms, Laura, 1991, et al.
(författare)
-
TSCH Meets BLE: Routed Mesh Communication Over BLE
- 2023
-
Ingår i: Proceedings - 19th International Conference on Distributed Computing in Smart Systems and the Internet of Things, DCOSS-IoT 2023. ; , s. 187-195
-
Konferensbidrag (refereegranskat)abstract
- Bluetooth Low Energy (BLE) is the prevalent communication protocol for the Internet of Things. However, for time-critical applications requiring time-synchronized multi-hop networks with often multiple node exchanging data at the same time slot, BLE lacks a solution. Instead, we commonly see IEEE 802.15.4 being used with its Time-Slotted Channel Hopping (TSCH) MAC layer. In this work, we build TBLE, which brings the established TSCH protocol to BLE, enabling BLE to be used for time-synchronized routed mesh communication. We show that in experimental testbed deployments, TBLE achieves similar performance to TSCH, with the possibility for lower average latencies of up to 20%. Moreover, due to the higher spectral efficiency of BLE compared with IEEE 802.15.4 (40 vs. 16 channels), more parallel routed communications are possible with TBLE, further reducing latency and increasing throughput.
|
|
| 9. |
- Hojjat, Ali, et al.
(författare)
-
ProgDTD: Progressive Learned Image Compression with Double-Tail-Drop Training
- 2023
-
Ingår i: IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops. - 2160-7516 .- 2160-7508. ; 2023-June, s. 1130-1139
-
Konferensbidrag (refereegranskat)abstract
- Progressive compression allows images to start loading as low-resolution versions, becoming clearer as more data is received. This increases user experience when, for example, network connections are slow. Today, most approaches for image compression, both classical and learned ones, are designed to be non-progressive. This paper introduces ProgDTD, a training method that transforms learned, non-progressive image compression approaches into progressive ones. The design of ProgDTD is based on the observation that the information stored within the bottleneck of a compression model commonly varies in importance. To create a progressive compression model, ProgDTD modifies the training steps to enforce the model to store the data in the bottleneck sorted by priority. We achieve progressive compression by transmitting the data in order of its sorted index. ProgDTD is designed for CNN-based learned image compression models, does not need additional parameters, and has a customizable range of progressiveness. For evaluation, we apply ProgDTD to the hyperprior model, one of the most common structures in learned image compression. Our experimental results show that ProgDTD performs comparably to its non-progressive counterparts and other state-of-the-art progressive models in terms of MS-SSIM and accuracy.
|
|
| 10. |
- Poirot, Valentin, 1994, et al.
(författare)
-
BlueSeer: AI-Driven Environment Detection via BLE Scans
- 2022
-
Ingår i: Proceedings - Design Automation Conference. - New York, NY, USA : ACM. - 0738-100X. ; , s. 871-876
-
Konferensbidrag (refereegranskat)abstract
- IoT devices rely on environment detection to trigger specific actions, e.g., for headphones to adapt noise cancellation to the surroundings. While phones feature many sensors, from GNSS to cameras, small wearables must rely on the few energy-efficient components they already incorporate. In this paper, we demonstrate that a Bluetooth radio is the only component required to accurately classify environments and present BlueSeer, an environment-detection system that solely relies on received BLE packets and an embedded neural network. BlueSeer achieves an accuracy of up to 84% differentiating between 7 environments on resource-constrained devices, and requires only ~12 ms for inference on a 64 MHz microcontroller-unit.
|
|
