Yixuan Gao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yixuan Gao has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Rajalakshmi Nandakumar
Cornell Tech, US, NYC, USA
8 shared publications - Thijs Roumen
Information Science, Cornell Tech, New York, New York, USA
5 shared publications - Zekun Chang
Information Science, Cornell Tech, New York, New York, USA
4 shared publications - Adnan Armouti
Cornell Tech, New York, USA
3 shared publications - Tanvir Ahmed
Cornell Tech, New York, USA
3 shared publications - Alexander T. Adams
School of Interactive Computing, Georgia Institute of Technology, Atlanta, Georgia, USA
2 shared publications
Latest Publications
- Open Access
- 2 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-12
StitchOver: Technical Embroidery on Seamed Fabrics
- 2 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessStitchOver: Technical Embroidery on Seamed Fabrics
- Open Access
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1311-1327
mmFHE: mmWave Sensing with End-to-End Fully Homomorphic Encryption
We present mmFHE, the first system that executes the entire cloud-side mmWave sensing pipeline including the DSP and ML inference under fully homomorphic encryption (FHE). mmFHE encrypts range profiles on an edge device after lightweight plaintext preprocessing and executes the entire mmWave signal-processing and ML inference pipeline homomorphically on a semi-honest cloud that operates exclusively on ciphertexts. At the core of mmFHE is a library of seven composable, data-oblivious FHE kernels that replace standard DSP routines with fixed arithmetic circuits for different application-specific pipelines. We demonstrate this approach on two representative tasks: vital-sign monitoring and gesture recognition. We formally prove two cryptographic guarantees for any pipeline assembled from this library: input privacy and data obliviousness. These guarantees effectively neutralize various supervised and unsupervised privacy attacks on raw data, including re-identification and data-dependent privacy leakage. Evaluation on three public radar datasets shows that encryption introduces negligible error versus the plaintext pipeline, with 84.5% gesture accuracy (vs. 84.7%). End-to-end cloud GPU latency is 1.21 s per 10 s vital-sign window and 5.76 s per 3 s gesture window. These results establish the initial feasibility of end-to-end mmWave sensing under FHE on commodity hardware.
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1311-1327
Open AccessmmFHE: mmWave Sensing with End-to-End Fully Homomorphic Encryption
We present mmFHE, the first system that executes the entire cloud-side mmWave sensing pipeline including the DSP and ML inference under fully homomorphic encryption (FHE). mmFHE encrypts range profiles on an edge device after lightweight plaintext preprocessing and executes the entire mmWave signal-processing and ML inference pipeline homomorphically on a semi-honest cloud that operates exclusively on ciphertexts. At the core of mmFHE is a library of seven composable, data-oblivious FHE kernels that replace standard DSP routines with fixed arithmetic circuits for different application-specific pipelines. We demonstrate this approach on two representative tasks: vital-sign monitoring and gesture recognition. We formally prove two cryptographic guarantees for any pipeline assembled from this library: input privacy and data obliviousness. These guarantees effectively neutralize various supervised and unsupervised privacy attacks on raw data, including re-identification and data-dependent privacy leakage. Evaluation on three public radar datasets shows that encryption introduces negligible error versus the plaintext pipeline, with 84.5% gesture accuracy (vs. 84.7%). End-to-end cloud GPU latency is 1.21 s per 10 s vital-sign window and 5.76 s per 3 s gesture window. These results establish the initial feasibility of end-to-end mmWave sensing under FHE on commodity hardware.
- Open Access
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1888-1890
Demo: mmFHE: mmWave Sensing with End-to-End Fully Homomorphic Encryption
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1888-1890
Open AccessDemo: mmFHE: mmWave Sensing with End-to-End Fully Homomorphic Encryption
- Open Access
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 71-76
RadarAgent: Agentic Tool Utilization for Radar-to-Language Understanding
MmWave radar senses human motion without recording appearance or identity, which makes it attractive for privacy-aware applications. However, its raw point clouds and signal-level features are hard for people to read. Existing radar-language models make such data queryable by training a radar tokenizer aligned to a language model, at the cost of specialized pipelines, paired data, and architecture-specific engineering. We present RadarAgent, an agentic framework that exposes radar processing as callable tools and lets a foundation model reason over them to describe motion in natural language, with no radar-token pretraining. In HumanML3D-style motion captioning, it exceeds the Radar-LLM baseline on key semantic metrics across several backbones, improving ROUGE-L from 36.0 to 39.6 and BERTScore from 83.3 to 90.7, and tool calling clearly outperforms flat feature prompting.
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 71-76
Open AccessRadarAgent: Agentic Tool Utilization for Radar-to-Language Understanding
MmWave radar senses human motion without recording appearance or identity, which makes it attractive for privacy-aware applications. However, its raw point clouds and signal-level features are hard for people to read. Existing radar-language models make such data queryable by training a radar tokenizer aligned to a language model, at the cost of specialized pipelines, paired data, and architecture-specific engineering. We present RadarAgent, an agentic framework that exposes radar processing as callable tools and lets a foundation model reason over them to describe motion in natural language, with no radar-token pretraining. In HumanML3D-style motion captioning, it exceeds the Radar-LLM baseline on key semantic metrics across several backbones, improving ROUGE-L from 36.0 to 39.6 and BERTScore from 83.3 to 90.7, and tool calling clearly outperforms flat feature prompting.
- 10 September 2026
- Published by Springer Nature
- p. 148-166
mmIR: Frequency-Space Inverse Rendering for 3D Millimeter-Wave Radar ADC Synthesis
- 10 September 2026
- Published by Springer Nature
- p. 148-166
mmIR: Frequency-Space Inverse Rendering for 3D Millimeter-Wave Radar ADC Synthesis
- Open Access
- 20 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Demonstrating OriStitch: A Machine Embroidery Workflow to Turn Existing Fabrics into Self-Folding 3D Textiles
- 20 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-2
Open AccessDemonstrating OriStitch: A Machine Embroidery Workflow to Turn Existing Fabrics into Self-Folding 3D Textiles
- Open Access
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-14
OriStitch: A Machine Embroidery Workflow to Turn Existing Fabrics into Self-Folding 3D Textiles
- 19 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-14
Open AccessOriStitch: A Machine Embroidery Workflow to Turn Existing Fabrics into Self-Folding 3D Textiles
- Open Access
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 19-24
Assembly Stethoscope: Detecting Assembly Errors through Frequency Sweeping-A Feasibility Study
- 4 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 19-24
Open AccessAssembly Stethoscope: Detecting Assembly Errors through Frequency Sweeping-A Feasibility Study
- Open Access
- 26 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 37-42
VitalHide: Enabling Privacy-Aware Wireless Sensing of Vital Signs
Wireless sensing technologies leveraging the acoustic and RF sensors in smart devices have advanced to enable accurate contactless monitoring of vital signs such as breathing and heartbeat. Wireless signal monitoring can enable important health applications, such as determining emotional states, detecting critical health conditions, and providing suitable interventions. While offering non-invasiveness and privacy advantages over wearables and cameras, these technologies pose serious privacy risks. The improved sensing accuracy allows any unauthorized sensor to detect sensitive health data without consent. This paper proposes VitalHide, a novel approach to protect vital signs from unauthorized sensing while still enabling wireless sensing for authorized devices. The key idea of VitalHide is to obfuscate the user's vital signs by generating a deceiving vital sign motion using a wearable mechanical device. The obfuscated signal can then only be decoded at authorized devices, while unauthorized devices receive deceiving vital signs. We show the feasibility of VitalHide by generating motion using a vibration sensor and shape memory alloy (SMA)-based smart textile, and the successful decoding of the obfuscated vital sign signal at the authorized receiver. This work advances privacy-preserving wireless sensing by securing personal health information against unauthorized access without affecting the utility of authorized devices.
- 26 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 37-42
Open AccessVitalHide: Enabling Privacy-Aware Wireless Sensing of Vital Signs
Wireless sensing technologies leveraging the acoustic and RF sensors in smart devices have advanced to enable accurate contactless monitoring of vital signs such as breathing and heartbeat. Wireless signal monitoring can enable important health applications, such as determining emotional states, detecting critical health conditions, and providing suitable interventions. While offering non-invasiveness and privacy advantages over wearables and cameras, these technologies pose serious privacy risks. The improved sensing accuracy allows any unauthorized sensor to detect sensitive health data without consent. This paper proposes VitalHide, a novel approach to protect vital signs from unauthorized sensing while still enabling wireless sensing for authorized devices. The key idea of VitalHide is to obfuscate the user's vital signs by generating a deceiving vital sign motion using a wearable mechanical device. The obfuscated signal can then only be decoded at authorized devices, while unauthorized devices receive deceiving vital signs. We show the feasibility of VitalHide by generating motion using a vibration sensor and shape memory alloy (SMA)-based smart textile, and the successful decoding of the obfuscated vital sign signal at the authorized receiver. This work advances privacy-preserving wireless sensing by securing personal health information against unauthorized access without affecting the utility of authorized devices.
- Open Access
- 18 September 2024
- Published by JMIR Publications Inc. in JMIRx Med
- Vol. 5 (1) , e65727
Peer Review of “All You Need Is Context: Clinician Evaluations of Various Iterations of a Large Language Model–Based First Aid Decision Support Tool in Ghana (Preprint)”
This is a peer review report related to MS ID#60477
- 18 September 2024
- Published by JMIR Publications Inc. in JMIRx Med
- Vol. 5 (1) , e65727
Open AccessPeer Review of “All You Need Is Context: Clinician Evaluations of Various Iterations of a Large Language Model–Based First Aid Decision Support Tool in Ghana (Preprint)”
This is a peer review report related to MS ID#60477