We present the first rigorous security, performance, energy, and cost analyses of the state-of-the-art on-DRAM-die read disturbance mitigation method, Per Row Activation Counting (PRAC), described in JEDEC DDR5 specification's April 2024 update. Unlike prior state-of-the-art that advises the memory controller to periodically issue refresh management (RFM) commands, which provides the DRAM chip with time to perform refreshes, PRAC introduces a new back-off signal. PRAC's back-off signal propagates from the DRAM chip to the memory controller and forces the memory controller to 1) stop serving requests and 2) issue RFM commands. As a result, RFM commands are issued when needed as opposed to periodically, reducing RFM's overheads. We analyze PRAC in four steps. First, we define an adversarial access pattern that represents the worst-case for PRAC's security. Second, we investigate PRAC's configurations and security implications. Our analyses show that PRAC can be configured for secure operation as long as no bitflip occurs before accessing a memory location 10 times. Third, we evaluate the performance impact of PRAC and compare it against prior works using Ramulator 2.0. Our analysis shows that while PRAC incurs less than 13% performance overhead for today's DRAM chips, its performance overheads can reach up to 94% for future DRAM chips that are more vulnerable to read disturbance bitflips. Fourth, we define an availability adversarial access pattern that exacerbates PRAC's performance overhead to perform a memory performance attack, demonstrating that such an adversarial pattern can hog up to 94% of DRAM throughput and degrade system throughput by up to 95%. We discuss PRAC's implications on future systems and foreshadow future research directions. To aid future research, we open-source our implementations and scripts at //github.com/CMU-SAFARI/ramulator2.
相關內容
We derive and study time-uniform confidence spheres -- confidence sphere sequences (CSSs) -- which contain the mean of random vectors with high probability simultaneously across all sample sizes. Our results include a dimension-free CSS for log-concave random vectors, a dimension-free CSS for sub-Gaussian random vectors, and CSSs for sub-$\psi$ random vectors (which includes sub-gamma, sub-Poisson, and sub-exponential distributions). For sub-Gaussian distributions we also provide a CSS which tracks a time-varying mean, generalizing Robbins' mixture approach to the multivariate setting. Finally, we provide several CSSs for heavy-tailed random vectors (two moments only). Our bounds hold under a martingale assumption on the mean and do not require that the observations be iid. Our work is based on PAC-Bayesian theory and inspired by an approach of Catoni and Giulini.
The rapid development and deployment of Generative AI in social settings raise important questions about how to optimally personalize them for users while maintaining accuracy and realism. Based on a Facebook public post-comment dataset, this study evaluates the ability of Llama 3.0 (70B) to predict the semantic tones across different combinations of a commenter's and poster's gender, age, and friendship closeness and to replicate these differences in LLM-generated comments. The study consists of two parts: Part I assesses differences in semantic tones across social relationship categories, and Part II examines the similarity between comments generated by Llama 3.0 (70B) and human comments from Part I given public Facebook posts as input. Part I results show that including social relationship information improves the ability of a model to predict the semantic tone of human comments. However, Part II results show that even without including social context information in the prompt, LLM-generated comments and human comments are equally sensitive to social context, suggesting that LLMs can comprehend semantics from the original post alone. When we include all social relationship information in the prompt, the similarity between human comments and LLM-generated comments decreases. This inconsistency may occur because LLMs did not include social context information as part of their training data. Together these results demonstrate the ability of LLMs to comprehend semantics from the original post and respond similarly to human comments, but also highlights their limitations in generalizing personalized comments through prompting alone.
Strategies for partially observable Markov decision processes (POMDP) typically require memory. One way to represent this memory is via automata. We present a method to learn an automaton representation of a strategy using a modification of the L*-algorithm. Compared to the tabular representation of a strategy, the resulting automaton is dramatically smaller and thus also more explainable. Moreover, in the learning process, our heuristics may even improve the strategy's performance. In contrast to approaches that synthesize an automaton directly from the POMDP thereby solving it, our approach is incomparably more scalable.
We present the first explicit construction of two-sided lossless expanders in the unbalanced setting (bipartite graphs that have many more nodes on the left than on the right). Prior to our work, all known explicit constructions in the unbalanced setting achieved only one-sided lossless expansion. Specifically, we show that the one-sided lossless expanders constructed by Kalev and Ta-Shma (RANDOM'22)--that are based on multiplicity codes introduced by Kopparty, Saraf, and Yekhanin (STOC'11)--are, in fact, two-sided lossless expanders. Using our unbalanced bipartite expander, we easily obtain lossless (non-bipartite) expander graphs on $N$ vertices with polynomial degree and a free group action. As far as we know, this is the first explicit construction of lossless (non-bipartite) expanders with $N$ vertices and degree $\ll N$.
Recognizing if LLM output can be grounded in evidence is central to many tasks in NLP: retrieval-augmented generation, summarization, document-grounded dialogue, and more. Current approaches to this kind of fact-checking are based on verifying each piece of a model generation against potential evidence using an LLM. However, this process can be very computationally expensive, requiring many calls to a model to check a single response. In this work, we show how to build small fact-checking models that have GPT-4-level performance but for 400x lower cost. We do this by constructing synthetic training data with GPT-4, which involves creating realistic yet challenging instances of factual errors via a structured generation procedure. Training on this data teaches models to check each fact in the claim and recognize synthesis of information across sentences. For evaluation, we unify datasets from recent work on fact-checking and grounding LLM generations into a new benchmark, LLM-AggreFact. Our best system MiniCheck-FT5 (770M parameters) outperforms all systems of comparable size and reaches GPT-4 accuracy. We release LLM-AggreFact, code for data synthesis, and models.
We give a novel nonparametric pointwise consistent statistical test (the Markov Checker) of the Markov condition for directed acyclic graph (DAG) or completed partially directed acyclic graph (CPDAG) models given a dataset. We also introduce the Cross-Algorithm Frugality Search (CAFS) for rejecting DAG models that either do not pass the Markov Checker test or that are not edge minimal. Edge minimality has been used previously by Raskutti and Uhler as a nonparametric simplicity criterion, though CAFS readily generalizes to other simplicity conditions. Reference to the ground truth is not necessary for CAFS, so it is useful for finding causal structure learning algorithms and tuning parameter settings that output causal models that are approximately true from a given data set. We provide a software tool for this analysis that is suitable for even quite large or dense models, provided a suitably fast pointwise consistent test of conditional independence is available. In addition, we show in simulation that the CAFS procedure can pick approximately correct models without knowing the ground truth.
Various kinds of uncertainty can occur in event logs, e.g., due to flawed recording, data quality issues, or the use of probabilistic models for activity recognition. Stochastically known event logs make these uncertainties transparent by encoding multiple possible realizations for events. However, the number of realizations encoded by a stochastically known log grows exponentially with its size, making exhaustive exploration infeasible even for moderately sized event logs. Thus, considering only the top-K most probable realizations has been proposed in the literature. In this paper, we implement an efficient algorithm to calculate a top-K realization ranking of an event log under event independence within O(Kn), where n is the number of uncertain events in the log. This algorithm is used to investigate the benefit of top-K rankings over top-1 interpretations of stochastically known event logs. Specifically, we analyze the usefulness of top-K rankings against different properties of the input data. We show that the benefit of a top-K ranking depends on the length of the input event log and the distribution of the event probabilities. The results highlight the potential of top-K rankings to enhance uncertainty-aware process mining techniques.
This study introduces a new framework for the artificial intelligence-assisted characterization of Gram-stained whole-slide images (WSIs). As a test for the diagnosis of bloodstream infections, Gram stains provide critical early data to inform patient treatment. Rapid and reliable analysis of Gram stains has been shown to be positively associated with better clinical outcomes, underscoring the need for improved tools to automate Gram stain analysis. In this work, we developed a novel transformer-based model for Gram-stained WSI classification, which is more scalable to large datasets than previous convolutional neural network (CNN) -based methods as it does not require patch-level manual annotations. We also introduce a large Gram stain dataset from Dartmouth-Hitchcock Medical Center (Lebanon, New Hampshire, USA) to evaluate our model, exploring the classification of five major categories of Gram-stained WSIs: Gram-positive cocci in clusters, Gram-positive cocci in pairs/chains, Gram-positive rods, Gram-negative rods, and slides with no bacteria. Our model achieves a classification accuracy of 0.858 (95% CI: 0.805, 0.905) and an AUC of 0.952 (95% CI: 0.922, 0.976) using five-fold nested cross-validation on our 475-slide dataset, demonstrating the potential of large-scale transformer models for Gram stain classification. We further demonstrate the generalizability of our trained model, which achieves strong performance on external datasets without additional fine-tuning.
The dominating NLP paradigm of training a strong neural predictor to perform one task on a specific dataset has led to state-of-the-art performance in a variety of applications (eg. sentiment classification, span-prediction based question answering or machine translation). However, it builds upon the assumption that the data distribution is stationary, ie. that the data is sampled from a fixed distribution both at training and test time. This way of training is inconsistent with how we as humans are able to learn from and operate within a constantly changing stream of information. Moreover, it is ill-adapted to real-world use cases where the data distribution is expected to shift over the course of a model's lifetime. The first goal of this thesis is to characterize the different forms this shift can take in the context of natural language processing, and propose benchmarks and evaluation metrics to measure its effect on current deep learning architectures. We then proceed to take steps to mitigate the effect of distributional shift on NLP models. To this end, we develop methods based on parametric reformulations of the distributionally robust optimization framework. Empirically, we demonstrate that these approaches yield more robust models as demonstrated on a selection of realistic problems. In the third and final part of this thesis, we explore ways of efficiently adapting existing models to new domains or tasks. Our contribution to this topic takes inspiration from information geometry to derive a new gradient update rule which alleviate catastrophic forgetting issues during adaptation.
This work considers the question of how convenient access to copious data impacts our ability to learn causal effects and relations. In what ways is learning causality in the era of big data different from -- or the same as -- the traditional one? To answer this question, this survey provides a comprehensive and structured review of both traditional and frontier methods in learning causality and relations along with the connections between causality and machine learning. This work points out on a case-by-case basis how big data facilitates, complicates, or motivates each approach.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191