Dissertation/ Thesis
Transparent Deployment of Machine Learning Models on Many-Accelerator Architectures
| Τίτλος: | Transparent Deployment of Machine Learning Models on Many-Accelerator Architectures |
|---|---|
| Συγγραφείς: | Chiu, Kuan-Lin |
| Έτος έκδοσης: | 2025 |
| Συλλογή: | Columbia University: Academic Commons |
| Θεματικοί όροι: | Computer science, Machine learning, Computer architecture, Systems on a chip--Design and construction, C++ (Computer program language), Computer software--Design and construction, Software engineering, Application software--Development |
| Περιγραφή: | The growing demand for machine learning (ML) and signal processing in embedded and edge systems introduces challenges in balancing performance, energy efficiency, and software development simplicity. Heterogeneous Systems-on-Chip (SoCs), which combine general-purpose processors with specialized accelerators, offer a promising solution. However, deploying applications efficiently on such architectures requires new hardware/software co-design methodologies, integration strategies, and runtime resource management.This dissertation presents an approach that separates application logic from accelerator integration, enabling more efficient deployment of ML and signal processing workloads on many-accelerator SoCs. To begin, I introduce ESP4ML, an open-source design flow that combines ESP, a modular SoC platform, with hls4ml, a high-level synthesis tool for generating ML accelerators. ESP4ML provides an embedded runtime software application programming interface (API) for managing accelerators dynamically within Linux. It enables data transfer between accelerators, reducing memory accesses.These features allow rapid prototyping of SoCs and are demonstrated through FPGA-based implementations executing end-to-end embedded workloads. Following that, I analyze communication models by comparing memory-based and point-to-point (p2p) data transfer mechanisms for accelerators. Using synthetic and real-world benchmarks, such as Nightvision and Wide-Area Motion Imagery (WAMI), I show that p2p communication consistently delivers better performance and energy efficiency, particularly in multi-threaded and tile-based accelerator systems. Next, a co-design strategy is presented that integrates the Eigen C++ linear algebra library with ESP. This enables high-level software to transparently offload computations to accelerators, achieving significant gains in both performance and energy efficiency without compromising software simplicity. To further improve the transparency of deploying ML workloads, I introduce WOLT, a lightweight ... |
| Τύπος εγγράφου: | thesis |
| Γλώσσα: | English |
| DOI: | 10.7916/qthh-b174 |
| Διαθεσιμότητα: | https://doi.org/10.7916/qthh-b174 |
| Αριθμός Καταχώρησης: | edsbas.CFB8DBA6 |
| Βάση Δεδομένων: | BASE |
| FullText | Text: Availability: 0 CustomLinks: – Url: https://doi.org/10.7916/qthh-b174# Name: EDS - BASE (ns324271) Category: fullText Text: View record from BASE |
|---|---|
| Header | DbId: edsbas DbLabel: BASE An: edsbas.CFB8DBA6 RelevancyScore: 892 AccessLevel: 3 PubType: Dissertation/ Thesis PubTypeId: dissertation PreciseRelevancyScore: 891.935607910156 |
| IllustrationInfo | |
| Items | – Name: Title Label: Title Group: Ti Data: Transparent Deployment of Machine Learning Models on Many-Accelerator Architectures – Name: Author Label: Authors Group: Au Data: <searchLink fieldCode="AR" term="%22Chiu%2C+Kuan-Lin%22">Chiu, Kuan-Lin</searchLink> – Name: DatePubCY Label: Publication Year Group: Date Data: 2025 – Name: Subset Label: Collection Group: HoldingsInfo Data: Columbia University: Academic Commons – Name: Subject Label: Subject Terms Group: Su Data: <searchLink fieldCode="DE" term="%22Computer+science%22">Computer science</searchLink><br /><searchLink fieldCode="DE" term="%22Machine+learning%22">Machine learning</searchLink><br /><searchLink fieldCode="DE" term="%22Computer+architecture%22">Computer architecture</searchLink><br /><searchLink fieldCode="DE" term="%22Systems+on+a+chip--Design+and+construction%22">Systems on a chip--Design and construction</searchLink><br /><searchLink fieldCode="DE" term="%22C%2B%2B+%28Computer+program+language%29%22">C++ (Computer program language)</searchLink><br /><searchLink fieldCode="DE" term="%22Computer+software--Design+and+construction%22">Computer software--Design and construction</searchLink><br /><searchLink fieldCode="DE" term="%22Software+engineering%22">Software engineering</searchLink><br /><searchLink fieldCode="DE" term="%22Application+software--Development%22">Application software--Development</searchLink> – Name: Abstract Label: Description Group: Ab Data: The growing demand for machine learning (ML) and signal processing in embedded and edge systems introduces challenges in balancing performance, energy efficiency, and software development simplicity. Heterogeneous Systems-on-Chip (SoCs), which combine general-purpose processors with specialized accelerators, offer a promising solution. However, deploying applications efficiently on such architectures requires new hardware/software co-design methodologies, integration strategies, and runtime resource management.This dissertation presents an approach that separates application logic from accelerator integration, enabling more efficient deployment of ML and signal processing workloads on many-accelerator SoCs. To begin, I introduce ESP4ML, an open-source design flow that combines ESP, a modular SoC platform, with hls4ml, a high-level synthesis tool for generating ML accelerators. ESP4ML provides an embedded runtime software application programming interface (API) for managing accelerators dynamically within Linux. It enables data transfer between accelerators, reducing memory accesses.These features allow rapid prototyping of SoCs and are demonstrated through FPGA-based implementations executing end-to-end embedded workloads. Following that, I analyze communication models by comparing memory-based and point-to-point (p2p) data transfer mechanisms for accelerators. Using synthetic and real-world benchmarks, such as Nightvision and Wide-Area Motion Imagery (WAMI), I show that p2p communication consistently delivers better performance and energy efficiency, particularly in multi-threaded and tile-based accelerator systems. Next, a co-design strategy is presented that integrates the Eigen C++ linear algebra library with ESP. This enables high-level software to transparently offload computations to accelerators, achieving significant gains in both performance and energy efficiency without compromising software simplicity. To further improve the transparency of deploying ML workloads, I introduce WOLT, a lightweight ... – Name: TypeDocument Label: Document Type Group: TypDoc Data: thesis – Name: Language Label: Language Group: Lang Data: English – Name: DOI Label: DOI Group: ID Data: 10.7916/qthh-b174 – Name: URL Label: Availability Group: URL Data: https://doi.org/10.7916/qthh-b174 – Name: AN Label: Accession Number Group: ID Data: edsbas.CFB8DBA6 |
| PLink | https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsbas&AN=edsbas.CFB8DBA6 |
| RecordInfo | BibRecord: BibEntity: Identifiers: – Type: doi Value: 10.7916/qthh-b174 Languages: – Text: English Subjects: – SubjectFull: Computer science Type: general – SubjectFull: Machine learning Type: general – SubjectFull: Computer architecture Type: general – SubjectFull: Systems on a chip--Design and construction Type: general – SubjectFull: C++ (Computer program language) Type: general – SubjectFull: Computer software--Design and construction Type: general – SubjectFull: Software engineering Type: general – SubjectFull: Application software--Development Type: general Titles: – TitleFull: Transparent Deployment of Machine Learning Models on Many-Accelerator Architectures Type: main BibRelationships: HasContributorRelationships: – PersonEntity: Name: NameFull: Chiu, Kuan-Lin IsPartOfRelationships: – BibEntity: Dates: – D: 01 M: 01 Type: published Y: 2025 Identifiers: – Type: issn-locals Value: edsbas – Type: issn-locals Value: edsbas.oa |
| ResultId | 1 |