arXiv subjects

G. Edward Suh

Publications and source records attributed to G. Edward Suh.

At least 37 records · Page 2Linked to original sources

arXiv2026

ReasoningBomb: A Stealthy Denial-of-Service Attack by Inducing Pathologically Long Reasoning in Large Reasoning Models

Large reasoning models (LRMs) extend large language models with explicit multi-step reasoning traces, but this capability introduces a new class of prompt-induced inference-time denial-of-service (PI-DoS) attacks that exploit the high computational cost of reasoning. We first formalize inference cost for LRMs and define PI-DoS, then prove that any practical PI-DoS attack should satisfy three properties: (1) a high amplification ratio, where each query induces a disproportionately long reasoning trace relative to its own length; (ii) stealthiness, in which prompts and responses remain on the natural language manifold and evade distribution shift detectors; and (iii) optimizability, in which the attack supports efficient optimization without being slowed by its own success. Under this framework, we present ReasoningBomb, a reinforcement-learning-based PI-DoS framework that is guided by a constant-time surrogate reward and trains a large reasoning-model attacker to generate short natural prompts that drive victim LRMs into pathologically long and often effectively non-terminating reasoning. Across seven open-source models (including LLMs and LRMs) and three commercial LRMs, ReasoningBomb induces 18,759 completion tokens on average and 19,263 reasoning tokens on average across reasoning models. It outperforms the the runner-up baseline by 35% in completion tokens and 38% in reasoning tokens, while inducing 6-7x more tokens than benign queries and achieving 286.7x input-to-output amplification ratio averaged across all samples. Additionally, our method achieves 99.8% bypass rate on input-based detection, 98.7% on output-based detection, and 98.4% against strict dual-stage joint detection.

cs.CR↗

arXiv2026

ReasAlign: Reasoning Enhanced Safety Alignment against Prompt Injection Attack

Large Language Models (LLMs) have enabled the development of powerful agentic systems capable of automating complex workflows across various fields. However, these systems are highly vulnerable to indirect prompt injection attacks, where malicious instructions embedded in external data can hijack agent behavior. In this work, we present ReasAlign, a model-level solution to improve safety alignment against indirect prompt injection attacks. The core idea of ReasAlign is to incorporate structured reasoning steps to analyze user queries, detect conflicting instructions, and preserve the continuity of the user's intended tasks to defend against indirect injection attacks. To further ensure reasoning logic and accuracy, we introduce a test-time scaling mechanism with a preference-optimized judge model that scores reasoning steps and selects the best trajectory. Comprehensive evaluations across various benchmarks show that ReasAlign maintains utility comparable to an undefended model while consistently outperforming Meta SecAlign, the strongest prior guardrail. On the representative open-ended CyberSecEval2 benchmark, which includes multiple prompt-injected tasks, ReasAlign achieves 94.6% utility and only 3.6% ASR, far surpassing the state-of-the-art defensive model of Meta SecAlign (56.4% utility and 74.4% ASR). These results demonstrate that ReasAlign achieves the best trade-off between security and utility, establishing a robust and practical defense against prompt injection attacks in real-world agentic systems. Our code and experimental results could be found at https://github.com/leolee99/ReasAlign.

cs.CR↗

arXiv2025

Machine Learning with Privacy for Protected Attributes

Differential privacy (DP) has become the standard for private data analysis. Certain machine learning applications only require privacy protection for specific protected attributes. Using naive variants of differential privacy in such use cases can result in unnecessary degradation of utility. In this work, we refine the definition of DP to create a more general and flexible framework that we call feature differential privacy (FDP). Our definition is simulation-based and allows for both addition/removal and replacement variants of privacy, and can handle arbitrary and adaptive separation of protected and non-protected features. We prove the properties of FDP, such as adaptive composition, and demonstrate its implications for limiting attribute inference attacks. We also propose a modification of the standard DP-SGD algorithm that satisfies FDP while leveraging desirable properties such as amplification via sub-sampling. We apply our framework to various machine learning tasks and show that it can significantly improve the utility of DP-trained models when public features are available. For example, we train diffusion models on the AFHQ dataset of animal faces and observe a drastic improvement in FID compared to DP, from 286.7 to 101.9 at $ε=8$, assuming that the blurred version of a training image is available as a public feature. Overall, our work provides a new approach to private data analysis that can help reduce the utility cost of DP while still providing strong privacy guarantees.

cs.CR↗

arXiv2025

How much do language models memorize?

We propose a new method for estimating how much a model knows about a datapoint and use it to measure the capacity of modern language models. Prior studies of language model memorization have struggled to disentangle memorization from generalization. We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we completely eliminate generalization, we can compute the total memorization, which provides an estimate of model capacity: our measurements estimate that GPT-style models have a capacity of approximately 3.6 bits per parameter. We train language models on datasets of increasing size and observe that models memorize until their capacity fills, at which point "grokking" begins, and unintended memorization decreases as models begin to generalize. We train hundreds of transformer language models ranging from $500K$ to $1.5B$ parameters and produce a series of scaling laws relating model capacity and data size to membership inference.

cs.CL↗

arXiv2025

Effectively Controlling Reasoning Models through Thinking Intervention

Reasoning-enhanced large language models (LLMs) explicitly generate intermediate reasoning steps prior to generating final answers, helping the model excel in complex problem-solving. In this paper, we demonstrate that this emerging generation framework offers a unique opportunity for more fine-grained control over model behavior. We propose Thinking Intervention, a novel paradigm designed to explicitly guide the internal reasoning processes of LLMs by strategically inserting or revising specific thinking tokens. We find that the Thinking Intervention paradigm enhances the capabilities of reasoning models across a wide range of tasks, including instruction following on IFEval and Overthinking, instruction hierarchy on SEP, and safety alignment on XSTest and SorryBench. Our results demonstrate that Thinking Intervention significantly outperforms baseline prompting approaches, achieving up to 6.7% accuracy gains in instruction-following scenarios, 15.4% improvements in reasoning about instruction hierarchies, and a 40.0% increase in refusal rates for unsafe prompts using open-source DeepSeek R1 models. Overall, our work opens a promising new research avenue for controlling reasoning LLMs.

cs.LG↗

arXiv2025

Sequence-Level Leakage Risk of Training Data in Large Language Models

This work quantifies the risk of training data leakage from LLMs (Large Language Models) using sequence-level probabilities. Computing extraction probabilities for individual sequences provides finer-grained information than has been studied in prior benchmarking work. We re-analyze the effects of decoding schemes, model sizes, prefix lengths, partial sequence leakages, and token positions to uncover new insights that were not possible in previous works due to their choice of metrics. We perform this study on two pre-trained models, Llama and OPT, trained on the Common Crawl and The Pile respectively. We discover that 1) Extraction Rate, the predominant metric used in prior quantification work, underestimates the threat of leakage of training data in randomized LLMs by as much as 2.14X. 2) Although on average, larger models and longer prefixes can extract more data, this is not true for a substantial portion of individual sequences. 30.4-41.5% of our sequences are easier to extract with either shorter prefixes or smaller models. 3) Contrary to previous beliefs, partial leakage in commonly used decoding schemes like top-k and top-p is not easier than leaking verbatim training data. The aim of this work is to encourage the adoption of this metric for future work on quantification of training data extraction.

cs.CL↗

arXiv2025

EcoServe: Designing Carbon-Aware AI Inference Systems

The rapid increase in LLM ubiquity and scale levies unprecedented demands on computing infrastructure. These demands not only incur large compute and memory resources but also significant energy, yielding large operational and embodied carbon emissions. In this work, we present three main observations based on modeling and traces from the production deployment of two Generative AI services in a major cloud service provider. First, while GPUs dominate operational carbon, host processing systems (e.g., CPUs, memory, storage) dominate embodied carbon. Second, offline, batch inference accounts for a significant portion (up to 55\%) of serving capacity. Third, there are different levels of heterogeneity across hardware and workloads for LLM inference. Based on these observations, we design EcoServe, a carbon-aware resource provision and scheduling framework for LLM serving systems. It is based on four principles - Reduce, Reuse, Rightsize, and Recycle (4R). With a cross-stack ILP formulation and design, we demonstrate that EcoServe can lower carbon emissions by up to 47\%, compared to performance, energy, and cost-optimized design points, while maintaining performance targets and SLOs.

cs.DC↗

arXiv2024

Towards Understanding Systems Trade-offs in Retrieval-Augmented Generation Model Inference

The rapid increase in the number of parameters in large language models (LLMs) has significantly increased the cost involved in fine-tuning and retraining LLMs, a necessity for keeping models up to date and improving accuracy. Retrieval-Augmented Generation (RAG) offers a promising approach to improving the capabilities and accuracy of LLMs without the necessity of retraining. Although RAG eliminates the need for continuous retraining to update model data, it incurs a trade-off in the form of slower model inference times. Resultingly, the use of RAG in enhancing the accuracy and capabilities of LLMs often involves diverse performance implications and trade-offs based on its design. In an effort to begin tackling and mitigating the performance penalties associated with RAG from a systems perspective, this paper introduces a detailed taxonomy and characterization of the different elements within the RAG ecosystem for LLMs that explore trade-offs within latency, throughput, and memory. Our study reveals underlying inefficiencies in RAG for systems deployment, that can result in TTFT latencies that are twice as long and unoptimized datastores that consume terabytes of storage.

cs.AR↗

arXiv2024

Information Flow Control in Machine Learning through Modular Model Architecture

In today's machine learning (ML) models, any part of the training data can affect the model output. This lack of control for information flow from training data to model output is a major obstacle in training models on sensitive data when access control only allows individual users to access a subset of data. To enable secure machine learning for access-controlled data, we propose the notion of information flow control for machine learning, and develop an extension to the Transformer language model architecture that strictly adheres to the IFC definition we propose. Our architecture controls information flow by limiting the influence of training data from each security domain to a single expert module, and only enables a subset of experts at inference time based on the access control policy.The evaluation using large text and code datasets show that our proposed parametric IFC architecture has minimal (1.9%) performance overhead and can significantly improve model accuracy (by 38% for the text dataset, and between 44%--62% for the code datasets) by enabling training on access-controlled data.

cs.LG↗

arXiv2024

Efficient Privacy-Preserving Machine Learning with Lightweight Trusted Hardware

In this paper, we propose a new secure machine learning inference platform assisted by a small dedicated security processor, which will be easier to protect and deploy compared to today's TEEs integrated into high-performance processors. Our platform provides three main advantages over the state-of-the-art: (i) We achieve significant performance improvements compared to state-of-the-art distributed Privacy-Preserving Machine Learning (PPML) protocols, with only a small security processor that is comparable to a discrete security chip such as the Trusted Platform Module (TPM) or on-chip security subsystems in SoCs similar to the Apple enclave processor. In the semi-honest setting with WAN/GPU, our scheme is 4X-63X faster than Falcon (PoPETs'21) and AriaNN (PoPETs'22) and 3.8X-12X more communication efficient. We achieve even higher performance improvements in the malicious setting. (ii) Our platform guarantees security with abort against malicious adversaries under honest majority assumption. (iii) Our technique is not limited by the size of secure memory in a TEE and can support high-capacity modern neural networks like ResNet18 and Transformer. While previous work investigated the use of high-performance TEEs in PPML, this work represents the first to show that even tiny secure hardware with really limited performance can be leveraged to significantly speed-up distributed PPML protocols if the protocol can be carefully designed for lightweight trusted hardware.

cs.CR↗

arXiv2024

LazyDP: Co-Designing Algorithm-Software for Scalable Training of Differentially Private Recommendation Models

Differential privacy (DP) is widely being employed in the industry as a practical standard for privacy protection. While private training of computer vision or natural language processing applications has been studied extensively, the computational challenges of training of recommender systems (RecSys) with DP have not been explored. In this work, we first present our detailed characterization of private RecSys training using DP-SGD, root-causing its several performance bottlenecks. Specifically, we identify DP-SGD's noise sampling and noisy gradient update stage to suffer from a severe compute and memory bandwidth limitation, respectively, causing significant performance overhead in training private RecSys. Based on these findings, we propose LazyDP, an algorithm-software co-design that addresses the compute and memory challenges of training RecSys with DP-SGD. Compared to a state-of-the-art DP-SGD training system, we demonstrate that LazyDP provides an average 119x training throughput improvement while also ensuring mathematically equivalent, differentially private RecSys models to be trained.

cs.IR↗

arXiv2023

GPU-based Private Information Retrieval for On-Device Machine Learning Inference

On-device machine learning (ML) inference can enable the use of private user data on user devices without revealing them to remote servers. However, a pure on-device solution to private ML inference is impractical for many applications that rely on embedding tables that are too large to be stored on-device. In particular, recommendation models typically use multiple embedding tables each on the order of 1-10 GBs of data, making them impractical to store on-device. To overcome this barrier, we propose the use of private information retrieval (PIR) to efficiently and privately retrieve embeddings from servers without sharing any private information. As off-the-shelf PIR algorithms are usually too computationally intensive to directly use for latency-sensitive inference tasks, we 1) propose novel GPU-based acceleration of PIR, and 2) co-design PIR with the downstream ML application to obtain further speedup. Our GPU acceleration strategy improves system throughput by more than $20 \times$ over an optimized CPU PIR implementation, and our PIR-ML co-design provides an over $5 \times$ additional throughput improvement at fixed model quality. Together, for various on-device ML applications such as recommendation and language modeling, our system on a single V100 GPU can serve up to $100,000$ queries per second -- a $>100 \times$ throughput improvement over a CPU-based baseline -- while maintaining model accuracy.

cs.CR↗

arXiv2023

Approximating ReLU on a Reduced Ring for Efficient MPC-based Private Inference

Secure multi-party computation (MPC) allows users to offload machine learning inference on untrusted servers without having to share their privacy-sensitive data. Despite their strong security properties, MPC-based private inference has not been widely adopted in the real world due to their high communication overhead. When evaluating ReLU layers, MPC protocols incur a significant amount of communication between the parties, making the end-to-end execution time multiple orders slower than its non-private counterpart. This paper presents HummingBird, an MPC framework that reduces the ReLU communication overhead significantly by using only a subset of the bits to evaluate ReLU on a smaller ring. Based on theoretical analyses, HummingBird identifies bits in the secret share that are not crucial for accuracy and excludes them during ReLU evaluation to reduce communication. With its efficient search engine, HummingBird discards 87--91% of the bits during ReLU and still maintains high accuracy. On a real MPC setup involving multiple servers, HummingBird achieves on average 2.03--2.67x end-to-end speedup without introducing any errors, and up to 8.64x average speedup when some amount of accuracy degradation can be tolerated, due to its up to 8.76x communication reduction.

cs.LG↗

arXiv2023

Bounding the Invertibility of Privacy-preserving Instance Encoding using Fisher Information

Privacy-preserving instance encoding aims to encode raw data as feature vectors without revealing their privacy-sensitive information. When designed properly, these encodings can be used for downstream ML applications such as training and inference with limited privacy risk. However, the vast majority of existing instance encoding schemes are based on heuristics and their privacy-preserving properties are only validated empirically against a limited set of attacks. In this paper, we propose a theoretically-principled measure for the privacy of instance encoding based on Fisher information. We show that our privacy measure is intuitive, easily applicable, and can be used to bound the invertibility of encodings both theoretically and empirically.

cs.LG↗

arXiv2023

AutoCAT: Reinforcement Learning for Automated Exploration of Cache-Timing Attacks

The aggressive performance optimizations in modern microprocessors can result in security vulnerabilities. For example, timing-based attacks in processor caches can steal secret keys or break randomization. So far, finding cache-timing vulnerabilities is mostly performed by human experts, which is inefficient and laborious. There is a need for automatic tools that can explore vulnerabilities given that unreported vulnerabilities leave the systems at risk. In this paper, we propose AutoCAT, an automated exploration framework that finds cache timing-channel attack sequences using reinforcement learning (RL). Specifically, AutoCAT formulates the cache timing-channel attack as a guessing game between an attack program and a victim program holding a secret. This guessing game can thus be solved via modern deep RL techniques. AutoCAT can explore attacks in various cache configurations without knowing design details and under different attack and victim program configurations. AutoCAT can also find attacks to bypass certain detection and defense mechanisms. In particular, AutoCAT discovered StealthyStreamline, a new attack that is able to bypass performance counter-based detection and has up to a 71% higher information leakage rate than the state-of-the-art LRU-based attacks on real processors. AutoCAT is the first of its kind in using RL for crafting microarchitectural timing-channel attack sequences and can accelerate cache timing-channel exploration for secure microprocessor designs.

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

DisaggRec: Architecting Disaggregated Systems for Large-Scale Personalized Recommendation

Deep learning-based personalized recommendation systems are widely used for online user-facing services in production datacenters, where a large amount of hardware resources are procured and managed to reliably provide low-latency services without disruption. As the recommendation models continue to evolve and grow in size, our analysis projects that datacenters deployed with monolithic servers will spend up to 12.4x total cost of ownership (TCO) to meet the requirement of model size and complexity over the next three years. Moreover, through in-depth characterization, we reveal that the monolithic server-based cluster suffers resource idleness and wastes up to 30% TCO by provisioning resources in fixed proportions. To address this challenge, we propose DisaggRec, a disaggregated system for large-scale recommendation serving. DisaggRec achieves the independent decoupled scaling-out of the compute and memory resources to match the changing demands from fast-evolving workloads. It also improves system reliability by segregating the failures of compute nodes and memory nodes. These two main benefits from disaggregation collectively reduce the TCO by up to 49.3%. Furthermore, disaggregation enables flexible and agile provisioning of increasing hardware heterogeneity in future datacenters. By deploying new hardware featuring near-memory processing capability, our evaluation shows that the disaggregated cluster achieves 21%-43.6% TCO savings over the monolithic server-based cluster across a three-year span of model evolution.

cs.DC↗

arXiv2022

Cocktail Party Attack: Breaking Aggregation-Based Privacy in Federated Learning using Independent Component Analysis

Federated learning (FL) aims to perform privacy-preserving machine learning on distributed data held by multiple data owners. To this end, FL requires the data owners to perform training locally and share the gradient updates (instead of the private inputs) with the central server, which are then securely aggregated over multiple data owners. Although aggregation by itself does not provably offer privacy protection, prior work showed that it may suffice if the batch size is sufficiently large. In this paper, we propose the Cocktail Party Attack (CPA) that, contrary to prior belief, is able to recover the private inputs from gradients aggregated over a very large batch size. CPA leverages the crucial insight that aggregate gradients from a fully connected layer is a linear combination of its inputs, which leads us to frame gradient inversion as a blind source separation (BSS) problem (informally called the cocktail party problem). We adapt independent component analysis (ICA)--a classic solution to the BSS problem--to recover private inputs for fully-connected and convolutional networks, and show that CPA significantly outperforms prior gradient inversion attacks, scales to ImageNet-sized inputs, and works on large batch sizes of up to 1024.

cs.LG↗