arXiv subjects

Murali Annavaram

Publications and source records attributed to Murali Annavaram.

At least 37 records · Page 2Linked to original sources

arXiv2025

Characterization of GPU TEE Overheads in Distributed Data Parallel ML Training

Confidential computing (CC) or trusted execution enclaves (TEEs) is now the most common approach to enable secure computing in the cloud. The recent introduction of GPU TEEs by NVIDIA enables machine learning (ML) models to be trained without leaking model weights or data to the cloud provider. However, the potential performance implications of using GPU TEEs for ML training are not well characterized. In this work, we present an in-depth characterization study on performance overhead associated with running distributed data parallel (DDP) ML training with GPU Trusted Execution Environments (TEE). Our study reveals the performance challenges in DDP training within GPU TEEs. DDP uses ring-all-reduce, a well-known approach, to aggregate gradients from multiple devices. Ring all-reduce consists of multiple scatter-reduce and all-gather operations. In GPU TEEs only the GPU package (GPU and HBM memory) is trusted. Hence, any data communicated outside the GPU packages must be encrypted and authenticated for confidentiality and integrity verification. Hence, each phase of the ring-all-reduce requires encryption and message authentication code (MAC) generation from the sender, and decryption and MAC authentication on the receiver. As the number of GPUs participating in DDP increases, the overhead of secure inter-GPU communication during ring-all-reduce grows proportionally. Additionally, larger models lead to more asynchronous all-reduce operations, exacerbating the communication cost. Our results show that with four GPU TEEs, depending on the model that is being trained, the runtime per training iteration increases by an average of 8x and up to a maximum of 41.6x compared to DDP training without TEE.

cs.CR↗

arXiv2025

DEL: Context-Aware Dynamic Exit Layer for Efficient Self-Speculative Decoding

Speculative Decoding (SD) is a widely used approach to accelerate the inference of large language models (LLMs) without reducing generation quality. It operates by first using a compact model to draft multiple tokens efficiently, followed by parallel verification using the target LLM. This approach leads to faster inference compared to auto-regressive decoding. While there are multiple approaches to create a draft model, one promising approach is to use early-exit methods. These methods draft candidate tokens by using a subset of layers of the primary model and applying the remaining layers for verification, allowing a single model to handle both drafting and verification. While this technique reduces memory usage and computational cost, its performance relies on the choice of the exit layer for drafting and the number of tokens drafted (speculation length) in each SD round. Prior works use hyperparameter exploration to statically select these values. However, our evaluations show that these hyperparameter values are task-specific, and even within a task they are dependent on the current sequence context. We introduce DEL (Dynamic Exit Layer), a plug-and-play method that adaptively selects the exit layer and speculation length during inference. DEL dynamically tracks the token acceptance rate if the tokens are drafted at each layer of an LLM and uses that knowledge to heuristically select the optimal exit layer and speculation length. Our experiments across a broad range of models and downstream tasks show that DEL achieves overall speedups of $2.16\times$$\sim$$2.62\times$ over vanilla auto-regressive decoding and improves upon state-of-the-art SD methods, which peak at $2.43\times$, by up to $0.19\times$. The code is available at https://github.com/hoenza/DEL.

cs.CL↗

arXiv2025

KVPR: Efficient LLM Inference with I/O-Aware KV Cache Partial Recomputation

Inference for Large Language Models (LLMs) is computationally demanding. To reduce the cost of auto-regressive decoding, Key-Value (KV) cache is used to store intermediate activations, which significantly lowers the computational overhead for token generation. However, the memory required for the KV cache grows rapidly, often exceeding the capacity of GPU memory. A cost-effective alternative is to offload KV cache to CPU memory, which alleviates GPU memory pressure, but shifts the bottleneck to the limited bandwidth of the PCIe connection between the CPU and GPU. Existing methods attempt to address these issues by overlapping GPU computation with I/O or employing CPU-GPU heterogeneous execution, but they are hindered by excessive data movement and dependence on CPU capabilities. Fully overlapping PCIe communication latency gets challenging as the size of the KV cache grows and/or the GPU compute capabilities increase. In this paper, we introduce KVPR, an efficient I/O-aware LLM inference method where the CPU first transfers a partial set of activations, from which the GPU can start recomputing the KV cache values. While the GPU recomputes the partial KV cache, the remaining portion of the KV cache is transferred concurrently from the CPU. This approach overlaps GPU recomputation with KV cache transfer to minimize idle GPU time and maximize inference performance. KVPR is fully automated by integrating a profiler module that utilizes input characteristics and system hardware information, a scheduler module to optimize the distribution of computation and communication workloads, and a runtime module to efficiently execute the derived execution plan. Experimental results show that KVPR achieves up to 35.8% lower latency and 46.2% higher throughput during decoding compared to state-of-the-art approaches. The code is available at https://github.com/chaoyij/KVPR.

cs.LG↗

arXiv2025

MARché: Fast Masked Autoregressive Image Generation with Cache-Aware Attention

Masked autoregressive (MAR) models unify the strengths of masked and autoregressive generation by predicting tokens in a fixed order using bidirectional attention for image generation. While effective, MAR models suffer from significant computational overhead, as they recompute attention and feed-forward representations for all tokens at every decoding step, despite most tokens remaining semantically stable across steps. We propose a training-free generation framework MARché to address this inefficiency through two key components: cache-aware attention and selective KV refresh. Cache-aware attention partitions tokens into active and cached sets, enabling separate computation paths that allow efficient reuse of previously computed key/value projections without compromising full-context modeling. But a cached token cannot be used indefinitely without recomputation due to the changing contextual information over multiple steps. MARché recognizes this challenge and applies a technique called selective KV refresh. Selective KV refresh identifies contextually relevant tokens based on attention scores from newly generated tokens and updates only those tokens that require recomputation, while preserving image generation quality. MARché significantly reduces redundant computation in MAR without modifying the underlying architecture. Empirically, MARché achieves up to 1.7x speedup with negligible impact on image quality, offering a scalable and broadly applicable solution for efficient masked transformer generation.

cs.LG↗

arXiv2025

High-Throughput Secure Multiparty Computation with an Honest Majority in Various Network Settings

In this work, we present novel protocols over rings for semi-honest secure three-party computation (3PC) and malicious four-party computation (4PC) with one corruption. While most existing works focus on improving total communication complexity, challenges such as network heterogeneity and computational complexity, which impact MPC performance in practice, remain underexplored. Our protocols address these issues by tolerating multiple arbitrarily weak network links between parties without any substantial decrease in performance. Additionally, they significantly reduce computational complexity by requiring up to half the number of basic instructions per gate compared to related work. These improvements lead to up to twice the throughput of state-of-the-art protocols in homogeneous network settings and up to eight times higher throughput in real-world heterogeneous settings. These advantages come at no additional cost: Our protocols maintain the best-known total communication complexity per multiplication, requiring 3 elements for 3PC and 5 elements for 4PC. We implemented our protocols alongside several state-of-the-art protocols (Replicated 3PC, ASTRA, Fantastic Four, Tetrad) in a novel open-source C++ framework optimized for high throughput. Five out of six implemented 3PC and 4PC protocols achieve more than one billion 32-bit multiplications or over 32 billion AND gates per second using our implementation in a 25 Gbit/s LAN environment. This represents the highest throughput achieved in 3PC and 4PC so far, outperforming existing frameworks like MP-SPDZ, ABY3, MPyC, and MOTION by two to three orders of magnitude.

cs.CR↗

arXiv2025

Differentially Private In-context Learning via Sampling Few-shot Mixed with Zero-shot Outputs

In-context learning (ICL) has shown promising improvement in downstream task adaptation of LLMs by augmenting prompts with relevant input-output examples (demonstrations). However, the ICL demonstrations can contain privacy-sensitive information, which can be leaked and/or regurgitated by the LLM output. Differential Privacy (DP), a widely adopted privacy safeguard, has emerged to mitigate this privacy leakage, with recent work demonstrating strong privacy-utility tradeoffs in classification tasks for ICL. However, generation tasks for ICL are challenging due to the high-dimensional output space of open-ended generation. To this end, we propose $\texttt{dps-mozo}$, Differentially Private Sampling by Mixing One-shot with Zero-shot Outputs, a decoding framework that generates DP text by sampling from the product of multiple one-shot outputs mixed with a zero-shot output. This mixing effectively reduces the amount of information that can be leaked by each demonstration. By utilizing the inherent randomness in sampling from the mixed distributions, we can achieve DP without adding noise, thereby improving the privacy-utility tradeoff. Our experimental evaluations show $\texttt{dps-mozo}$ can achieve a strong privacy guarantee, $ε=2$, with minimal utility degradation compared to non-private few-shot learning, $\textbf{0.3}$% ROUGE-L F1 score decrease on the SAMSum dataset with Gemma 2 2B.

cs.LG↗

arXiv2024

Fastrack: Fast IO for Secure ML using GPU TEEs

As cloud-based ML expands, ensuring data security during training and inference is critical. GPU-based Trusted Execution Environments (TEEs) offer secure, high-performance solutions, with CPU TEEs managing data movement and GPU TEEs handling authentication and computation. However, CPU-to-GPU communication overheads significantly hinder performance, as data must be encrypted, authenticated, decrypted, and verified, increasing costs by 12.69 to 33.53 times. This results in GPU TEE inference becoming 54.12% to 903.9% slower and training 10% to 455% slower than non-TEE systems, undermining GPU TEE advantages in latency-sensitive applications. This paper analyzes Nvidia H100 TEE protocols and identifies three key overheads: 1) redundant CPU re-encryption, 2) limited authentication parallelism, and 3) unnecessary operation serialization. We propose Fastrack, optimizing with 1) direct GPU TEE communication, 2) parallelized authentication, and 3) overlapping decryption with PCI-e transmission. These optimizations cut communication costs and reduce inference/training runtime by up to 84.6%, with minimal overhead compared to non-TEE systems.

cs.CR↗

arXiv2024

Adaptively Private Next-Token Prediction of Large Language Models

As Large Language Models (LLMs) proliferate, developing privacy safeguards for these models is crucial. One popular safeguard involves training LLMs in a differentially private manner. However, such solutions are shown to be computationally expensive and detrimental to the utility of these models. Since LLMs are deployed on the cloud and thus only accessible via an API, a Machine Learning as a Service (MLaaS) provider can protect its downstream data by privatizing the predictions during the decoding process. However, the practicality of such solutions still largely lags behind DP training methods. One recent promising approach, Private Mixing of Ensemble Distributions (PMixED), avoids additive noise by sampling from the output distributions of private LLMs mixed with the output distribution of a public model. Yet, PMixED must satisfy a fixed privacy level for a given number of queries, which is difficult for an analyst to estimate before inference and, hence, does not scale. To this end, we relax the requirements to a more practical setting by introducing Adaptive PMixED (AdaPMixED), a private decoding framework based on PMixED that is adaptive to the private and public output distributions evaluated on a given input query. In this setting, we introduce a noisy screening mechanism that filters out queries with potentially expensive privacy loss, and a data-dependent analysis that exploits the divergence of the private and public output distributions in its privacy loss calculation. Our experimental evaluations demonstrate that our mechanism and analysis can reduce the privacy loss by 16x while preserving the utility over the original PMixED. Furthermore, performing 100K predictions with AdaPMixED still achieves strong utility and a reasonable data-dependent privacy loss of 5.25.

cs.LG↗

arXiv2024

MPC-Pipe: an Efficient Pipeline Scheme for Secure Multi-party Machine Learning Inference

Multi-party computing (MPC) has been gaining popularity as a secure computing model over the past few years. However, prior works have demonstrated that MPC protocols still pay substantial performance penalties compared to plaintext, particularly when applied to ML algorithms. The overhead is due to added computation and communication costs. Prior studies, as well as our own analysis, found that most MPC protocols today sequentially perform communication and computation. The participating parties must compute on their shares first and then perform data communication to allow the distribution of new secret shares before proceeding to the next computation step. In this work, we show that serialization is unnecessary, particularly in the context of ML computations (both in Convolutional neural networks and in Transformer-based models). We demonstrate that it is possible to carefully orchestrate the computation and communication steps to overlap. We propose MPC-Pipe, an efficient MPC system for both training and inference of ML workloads, which pipelines computations and communications in an MPC protocol during the online phase. MPC-Pipe proposes three pipeline schemes to optimize the online phase of ML in the semi-honest majority adversary setting. We implement MPC-Pipe by augmenting a modified version of CrypTen, which separates online and offline phases. We evaluate the end-to-end system performance benefits of the online phase of MPC using deep neural networks (VGG16, ResNet50) and Transformers using different network settings. We show that MPC-Pipe can improve the throughput and latency of ML workloads.

cs.CR↗

arXiv2024

CADC: Encoding User-Item Interactions for Compressing Recommendation Model Training Data

Deep learning recommendation models (DLRMs) are at the heart of the current e-commerce industry. However, the amount of training data used to train these large models is growing exponentially, leading to substantial training hurdles. The training dataset contains two primary types of information: content-based information (features of users and items) and collaborative information (interactions between users and items). One approach to reduce the training dataset is to remove user-item interactions. But that significantly diminishes collaborative information, which is crucial for maintaining accuracy due to its inclusion of interaction histories. This loss profoundly impacts DLRM performance. This paper makes an important observation that if one can capture the user-item interaction history to enrich the user and item embeddings, then the interaction history can be compressed without losing model accuracy. Thus, this work, Collaborative Aware Data Compression (CADC), takes a two-step approach to training dataset compression. In the first step, we use matrix factorization of the user-item interaction matrix to create a novel embedding representation for both the users and items. Once the user and item embeddings are enriched by the interaction history information the approach then applies uniform random sampling of the training dataset to drastically reduce the training dataset size while minimizing model accuracy drop. The source code of CADC is available at \href{https://anonymous.4open.science/r/DSS-RM-8C1D/README.md}{https://anonymous.4open.science/r/DSS-RM-8C1D/README.md}.

cs.IR↗

arXiv2024

Differentially Private Next-Token Prediction of Large Language Models

Ensuring the privacy of Large Language Models (LLMs) is becoming increasingly important. The most widely adopted technique to accomplish this is DP-SGD, which trains a model to guarantee Differential Privacy (DP). However, DP-SGD overestimates an adversary's capabilities in having white box access to the model and, as a result, causes longer training times and larger memory usage than SGD. On the other hand, commercial LLM deployments are predominantly cloud-based; hence, adversarial access to LLMs is black-box. Motivated by these observations, we present Private Mixing of Ensemble Distributions (PMixED): a private prediction protocol for next-token prediction that utilizes the inherent stochasticity of next-token sampling and a public model to achieve Differential Privacy. We formalize this by introducing RD-mollifers which project each of the model's output distribution from an ensemble of fine-tuned LLMs onto a set around a public LLM's output distribution, then average the projected distributions and sample from it. Unlike DP-SGD which needs to consider the model architecture during training, PMixED is model agnostic, which makes PMixED a very appealing solution for current deployments. Our results show that PMixED achieves a stronger privacy guarantee than sample-level privacy and outperforms DP-SGD for privacy $ε= 8$ on large-scale datasets. Thus, PMixED offers a practical alternative to DP training methods for achieving strong generative utility without compromising privacy.

cs.CR↗

arXiv2024

Ethos: Rectifying Language Models in Orthogonal Parameter Space

Language models (LMs) have greatly propelled the research on natural language processing. However, LMs also raise concerns regarding the generation of biased or toxic content and the potential disclosure of private information from the training dataset. In this work, we present a new efficient approach, Ethos, that rectifies LMs to mitigate toxicity and bias in outputs and avoid privacy leakage. Ethos is built on task arithmetic. However, unlike current task arithmetic algorithms, Ethos distinguishes general beneficial and undesired knowledge when reconstructing task vectors. Specifically, Ethos first obtains a set of principal components from the pre-trained models using singular value decomposition. Then, by projecting the task vector onto principal components, Ethos identifies the principal components that encode general or undesired knowledge. Ethos performs negating using the task vector with undesired knowledge only, thereby minimizing collateral damage on general model utility. We demonstrate the efficacy of our approach on three different tasks: debiasing, detoxification, and memorization unlearning. Evaluations show Ethos is more effective in removing undesired knowledge and maintaining the overall model performance compared to current task arithmetic methods.

cs.CL↗

arXiv2024

Edge Private Graph Neural Networks with Singular Value Perturbation

Graph neural networks (GNNs) play a key role in learning representations from graph-structured data and are demonstrated to be useful in many applications. However, the GNN training pipeline has been shown to be vulnerable to node feature leakage and edge extraction attacks. This paper investigates a scenario where an attacker aims to recover private edge information from a trained GNN model. Previous studies have employed differential privacy (DP) to add noise directly to the adjacency matrix or a compact graph representation. The added perturbations cause the graph structure to be substantially morphed, reducing the model utility. We propose a new privacy-preserving GNN training algorithm, Eclipse, that maintains good model utility while providing strong privacy protection on edges. Eclipse is based on two key observations. First, adjacency matrices in graph structures exhibit low-rank behavior. Thus, Eclipse trains GNNs with a low-rank format of the graph via singular values decomposition (SVD), rather than the original graph. Using the low-rank format, Eclipse preserves the primary graph topology and removes the remaining residual edges. Eclipse adds noise to the low-rank singular values instead of the entire graph, thereby preserving the graph privacy while still maintaining enough of the graph structure to maintain model utility. We theoretically show Eclipse provide formal DP guarantee on edges. Experiments on benchmark graph datasets show that Eclipse achieves significantly better privacy-utility tradeoff compared to existing privacy-preserving GNN training methods. In particular, under strong privacy constraints ($ε$ < 4), Eclipse shows significant gains in the model utility by up to 46%. We further demonstrate that Eclipse also has better resilience against common edge attacks (e.g., LPA), lowering the attack AUC by up to 5% compared to other state-of-the-art baselines.

cs.LG↗

arXiv2023

CompactTag: Minimizing Computation Overheads in Actively-Secure MPC for Deep Neural Networks

Secure Multiparty Computation (MPC) protocols enable secure evaluation of a circuit by several parties, even in the presence of an adversary who maliciously corrupts all but one of the parties. These MPC protocols are constructed using the well-known secret-sharing-based paradigm (SPDZ and SPDZ2k), where the protocols ensure security against a malicious adversary by computing Message Authentication Code (MAC) tags on the input shares and then evaluating the circuit with these input shares and tags. However, this tag computation adds a significant runtime overhead, particularly for machine learning (ML) applications with numerous linear computation layers such as convolutions and fully connected layers. To alleviate the tag computation overhead, we introduce CompactTag, a lightweight algorithm for generating MAC tags specifically tailored for linear layers in ML. Linear layer operations in ML, including convolutions, can be transformed into Toeplitz matrix multiplications. For the multiplication of two matrices with dimensions T1 x T2 and T2 x T3 respectively, SPDZ2k required O(T1 x T2 x T3) local multiplications for the tag computation. In contrast, CompactTag only requires O(T1 x T2 + T1 x T3 + T2 x T3) local multiplications, resulting in a substantial performance boost for various ML models. We empirically compared our protocol to the SPDZ2k protocol for various ML circuits, including ResNet Training-Inference, Transformer Training-Inference, and VGG16 Training-Inference. SPDZ2k dedicated around 30% of its online runtime for tag computation. CompactTag speeds up this tag computation bottleneck by up to 23x, resulting in up to 1.47x total online phase runtime speedups for various ML workloads.

cs.CR↗

arXiv2022

Attribute Inference Attack of Speech Emotion Recognition in Federated Learning Settings

Speech emotion recognition (SER) processes speech signals to detect and characterize expressed perceived emotions. Many SER application systems often acquire and transmit speech data collected at the client-side to remote cloud platforms for inference and decision making. However, speech data carry rich information not only about emotions conveyed in vocal expressions, but also other sensitive demographic traits such as gender, age and language background. Consequently, it is desirable for SER systems to have the ability to classify emotion constructs while preventing unintended/improper inferences of sensitive and demographic information. Federated learning (FL) is a distributed machine learning paradigm that coordinates clients to train a model collaboratively without sharing their local data. This training approach appears secure and can improve privacy for SER. However, recent works have demonstrated that FL approaches are still vulnerable to various privacy attacks like reconstruction attacks and membership inference attacks. Although most of these have focused on computer vision applications, such information leakages exist in the SER systems trained using the FL technique. To assess the information leakage of SER systems trained using FL, we propose an attribute inference attack framework that infers sensitive attribute information of the clients from shared gradients or model parameters, corresponding to the FedSGD and the FedAvg training algorithms, respectively. As a use case, we empirically evaluate our approach for predicting the client's gender information using three SER benchmark datasets: IEMOCAP, CREMA-D, and MSP-Improv. We show that the attribute inference attack is achievable for SER systems trained using FL. We further identify that most information leakage possibly comes from the first layer in the SER model.

cs.CR↗

arXiv2022

Data Leakage via Access Patterns of Sparse Features in Deep Learning-based Recommendation Systems

Online personalized recommendation services are generally hosted in the cloud where users query the cloud-based model to receive recommended input such as merchandise of interest or news feed. State-of-the-art recommendation models rely on sparse and dense features to represent users' profile information and the items they interact with. Although sparse features account for 99% of the total model size, there was not enough attention paid to the potential information leakage through sparse features. These sparse features are employed to track users' behavior, e.g., their click history, object interactions, etc., potentially carrying each user's private information. Sparse features are represented as learned embedding vectors that are stored in large tables, and personalized recommendation is performed by using a specific user's sparse feature to index through the tables. Even with recently-proposed methods that hides the computation happening in the cloud, an attacker in the cloud may be able to still track the access patterns to the embedding tables. This paper explores the private information that may be learned by tracking a recommendation model's sparse feature access patterns. We first characterize the types of attacks that can be carried out on sparse features in recommendation models in an untrusted cloud, followed by a demonstration of how each of these attacks leads to extracting users' private information or tracking users by their behavior over time.

cs.CE↗

arXiv2022

Secure and Fault Tolerant Decentralized Learning

Federated learning (FL) is a promising paradigm for training a global model over data distributed across multiple data owners without centralizing clients' raw data. However, sharing of local model updates can also reveal information of clients' local datasets. Trusted execution environments (TEEs) within the FL server have been recently deployed by companies like Meta for secure aggregation. However, secure aggregation can suffer from error-prone local updates sent by clients that become faulty during training due to underlying device malfunctions. Also, data heterogeneity across clients makes fault mitigation challenging, as even updates from normal clients are dissimilar. Thus, most of the prior fault tolerant methods, which treat any local update differing from the majority of other updates as faulty, perform poorly. We propose DiverseFL to make model aggregation secure as well as robust to faults. In DiverseFL, any client whose local model update diverges from its associated guiding update is tagged as being faulty. To implement our novel per-client criteria for fault mitigation, DiverseFL creates a TEE-based secure enclave within the FL server, which in addition to performing secure aggregation for carrying out the global model update step, securely receives a small representative sample of local data from each client only once before training, and computes guiding updates for each participating client during training. Thus, DiverseFL provides security against privacy leakage as well as robustness against faulty clients. In experiments, DiverseFL consistently achieves significant improvements in absolute test accuracy over prior fault mitigation benchmarks. DiverseFL also performs closely to OracleSGD, where server combines updates only from the normal clients. We also analyze the convergence rate of DiverseFL under non-IID data and standard convexity assumptions.

cs.DC↗

arXiv2022

LAORAM: A Look Ahead ORAM Architecture for Training Large Embedding Tables

Data confidentiality is becoming a significant concern, especially in the cloud computing era. Memory access patterns have been demonstrated to leak critical information such as security keys and a program's spatial and temporal information. This information leak poses an even more significant privacy challenge in machine learning models with embedding tables. Embedding tables are routinely used to learn categorical features from training data. Even knowing the locations of the embedding table entries accessed, not the data within the embedding table, will compromise categorical input data to the model. Embedding entries are privacy-sensitive since they disclose valuable properties about the user. Oblivious RAM (ORAM), and its enhanced variants such as PathORAM have emerged as viable solutions to hide leakage from memory access streams. In this work, we present LAORAM, an ORAM framework explicitly designed to protect user privacy during embedding table training. LAORAM exploits the unique property of training, the training samples used in the future are known beforehand. LAORAM preprocesses the training samples to identify the memory blocks which are accessed together in the near future. The system tries to assign these blocks to as few paths as possible within the PathORAM infrastructure. LAORAM does this operation by combining multiple blocks accessed together as superblocks. To further increase performance, LAORAM uses a fat-tree structure for PathORAM reducing the number of background evictions required, which improves the stash usage. We have evaluated LAORAM using both a recommendation model (DLRM) and a NLP model (XLM-R) embedding table configurations. LAORAM performs 5 times faster than PathORAM on a recommendation dataset (Kaggle) and 5.4x faster on a NLP dataset (XNLI), while guaranteeing the same security guarantees as the original PathORAM.

cs.CR↗