arXiv subjects

Danella Zhao

Publications and source records attributed to Danella Zhao.

6 recordsLinked to original sources

arXiv2026

Anchor: Augmented Cross-Layer Recurrent State for Long-Context Retrieval in Test-Time Training

Parameter-efficient linearization turns a pretrained autoregressive Transformer into an efficient model by hybridizing sliding-window attention (SWA) with a constant-size recurrent memory. But does every layer's recurrent state contribute to long-context retrieval? Investigating an SWA and Test-Time Training (TTT) hybrid, we find that retrieval is sparse across depth, concentrated in a few anchor layers, and that the remaining states hold similar information expressed through layer-specific transformations. We exploit both observations in Anchor: TTT memory placed only at the anchor layers, and a cross-layer recurrent state written once and read by many layers, each through its own learned read map that realizes a layer-specific stretch and rotation, along with demonstrated generalization to other hybrid attentions. On Llama-3.1-8B and Mistral-7B, Anchor delivers 2.3-3.1x higher recall and 6.7-8.9 points higher MMLU accuracy than LoLCATs with a 10x smaller recurrent state, retains 91-95% of the teacher's MMLU accuracy, and cuts end-to-end latency by up to 4.3x relative to softmax attention.

cs.LG↗

arXiv2026

Interflow: Breaking the Offline-Online Barrier for Private AI at the Edge

Privacy-preserving machine learning (PPML) offers a promising path toward bringing powerful AI services to the edge without exposing sensitive user data. However, its practical deployment remains constrained by the high communication, computation, and data-movement costs of secure multi-party computation (MPC), particularly during nonlinear inference. Existing MPC systems typically decouple input-independent correlated randomness generation from online execution, creating substantial memory pressure to buffer single-use cryptographic data. Moreover, limited storage can force a stop-and-refill execution pattern that repeatedly stalls the pipeline to replenish these buffers, bringing randomness generation back onto the critical path. We propose Interflow, an algorithm-hardware co-design that rethinks this separation by batch-wise processing and pipelining correlated-randomness generation with online execution: while the current batch consumes pre-generated randomness to execute secure nonlinear operations, correlation generation proceeds concurrently for the next batch. By leveraging interleaved execution, efficient data movement, and hardware acceleration, the design transforms preprocessing from a serialized, storage-intensive offline phase into a continuously overlapped pipeline that generates correlated randomness closer to the point of consumption, reducing memory footprint and data movement overhead. Implemented on the Zynq UltraScale+ ZCU102, Interflow improves nonlinear evaluation throughput by up to 1.55x for CNN operators and 1.48x for transformer operators. It achieves up to 1.26x speedup for ResNet-50 and up to 1.21x speedup for BERT-base, and reduces memory footprint by up to 37%, compared with baseline frameworks.

cs.AR↗

arXiv2026

TAMI-MPC:Trusted Acceleration of Minimal-Interaction MPC for Efficient Nonlinear Inference

Secure multi-party computation (MPC) offers a practical foundation for privacy-preserving machine learning at the edge. However, current MPC systems rely heavily on communication and computation-intensive primitives-such as secure comparison for nonlinear inference, which are often impractical on resource-constrained platforms. To enable real-time inference under a resource-constrained platform, we introduce a Trusted Acceleration of Minimal-Interaction MPC framework, TAMI-MPC, for nonlinear evaluation. Specifically, we reduce communication cost by redesigning the core primitives, leaf comparison, and tree merge, reducing the interactive round from log(n) to just 1 per operation. Furthermore, unlike prior work that heavily relies on oblivious transfer (OT), a well-known computational bottleneck, we leverage synchronized seeds inside the TEE to eliminate OT for the vast majority of our designs, along with a correlated-randomness reuse technique that keeps new designs computationally lightweight. To fully realize the potential, we design a specialized accelerator that restructures the dataflow across stages to enable continuous, fine-grained streaming and high parallelism, reducing memory overhead. Our design achieves up to 4.86x speedup on ResNet-50 inference, compared with state-of-the-art CNN frameworks, and achieves up to 7.44x speedup on BERT-base inference, compared with state-of-the-art LLM frameworks.

cs.AR↗

arXiv2026

SecDTD: Dynamic Token Drop for Secure Transformers Inference

The rapid adoption of Transformer-based AI has been driven by accessible models such as ChatGPT, which provide API-based services for developers and businesses. However, as these online inference services increasingly handle sensitive inputs, privacy concerns have emerged as a significant challenge. To address this, secure inference frameworks have been proposed, but their high computational and communication overhead often limit practical deployment. In plaintext settings, token drop is an effective technique for reducing inference cost; however, our analysis reveals that directly applying such methods to ciphertext scenarios is suboptimal due to distinct cost distributions in secure computation. We propose SecDTD, a dynamic token drop scheme tailored for secure Transformer inference. SecDTD advances token drop by shifting the dropping to earlier inference stages, effectively reducing the cost of key components such as Softmax. To support this, we introduce two core techniques. Max-Centric Normalization (MCN): A novel, Softmax-independent scoring method that enables early token drop with minimal overhead and improved normalization, supporting more aggressive dropping without accuracy loss. OMSel: A faster, oblivious median selection protocol that securely identifies the median of importance scores to support token drop. Compared to existing sorting-based methods, OMSel achieves a 16.9$\times$ speedup while maintaining security, obliviousness and randomness. We evaluate SecDTD through 48 experiments across eight GLUE datasets under various network settings using the BOLT and BumbleBee frameworks. SecDTD achieves 4.47 times end-to-end inference acceleration without degradation in accuracy.

cs.CR↗

arXiv2025

Silentflow: Leveraging Trusted Execution for Resource-Limited MPC via Hardware-Algorithm Co-design

Secure Multi-Party Computation (MPC) offers a practical foundation for privacy-preserving machine learning at the edge, with MPC commonly employed to support nonlinear operations. These MPC protocols fundamentally rely on Oblivious Transfer (OT), particularly Correlated OT (COT), to generate correlated randomness essential for secure computation. Although COT generation is efficient in conventional two-party settings with resource-rich participants, it becomes a critical bottleneck in real-world inference on resource-constrained devices (e.g., IoT sensors and wearables), due to both communication latency and limited computational capacity. To enable real-time secure inference, we introduce Silentflow, a highly efficient Trusted Execution Environment (TEE)-assisted protocol that eliminates communication in COT generation. We tackle the core performance bottleneck-low computational intensity-through structured algorithmic decomposition: kernel fusion for parallelism, Blocked On-chip eXpansion (BOX) to improve memory access patterns, and vectorized batch operations to maximize memory bandwidth utilization. Through design space exploration, we balance end-to-end latency and resource demands, achieving up to 39.51x speedup over state-of-the-art protocols. By offloading COT computations to a Zynq-7000 SoC, SilentFlow accelerates PPMLaaS inference on the ImageNet dataset under resource constraints, achieving a 4.62x and 3.95x speedup over Cryptflow2 and Cheetah, respectively.

cs.CR↗

arXiv2022

DeepAuditor: Distributed Online Intrusion Detection System for IoT devices via Power Side-channel Auditing

As the number of IoT devices has increased rapidly, IoT botnets have exploited the vulnerabilities of IoT devices. However, it is still challenging to detect the initial intrusion on IoT devices prior to massive attacks. Recent studies have utilized power side-channel information to identify this intrusion behavior on IoT devices but still lack accurate models in real-time for ubiquitous botnet detection. We proposed the first online intrusion detection system called DeepAuditor for IoT devices via power auditing. To develop the real-time system, we proposed a lightweight power auditing device called Power Auditor. We also designed a distributed CNN classifier for online inference in a laboratory setting. In order to protect data leakage and reduce networking redundancy, we then proposed a privacy-preserved inference protocol via Packed Homomorphic Encryption and a sliding window protocol in our system. The classification accuracy and processing time were measured, and the proposed classifier outperformed a baseline classifier, especially against unseen patterns. We also demonstrated that the distributed CNN design is secure against any distributed components. Overall, the measurements were shown to the feasibility of our real-time distributed system for intrusion detection on IoT devices.

cs.CR↗