搜索结果：Alter

Large language models (LLMs) have advanced to encompass extensive knowledge across diverse domains. Yet controlling what a LLMs should not know is important for ensuring alignment and thus safe use. However, effective unlearning in LLMs is difficult due to the fuzzy boundary between knowledge retention and forgetting. This challenge is exacerbated by entangled parameter spaces from continuous multi-domain training, often resulting in collateral damage, especially under aggressive unlearning strategies. Furthermore, the computational overhead required to optimize State-of-the-Art (SOTA) models with billions of parameters poses an additional barrier. In this work, we present ALTER, a lightweight unlearning framework for LLMs to address both the challenges of knowledge entanglement and unlearning efficiency. ALTER operates through two phases: (I) high entropy tokens are captured and learned via the shared A matrix in LoRA, followed by (II) an asymmetric LoRA architecture that achieves a specified forgetting objective by parameter isolation and unlearning tokens within the target subdomains. Serving as a new research direction for achieving unlearning via token-level isolation in the a

Alterations in historical manuscripts such as letters represent a promising field of research. On the one hand, they help understand the construction of text. On the other hand, topics that are being considered sensitive at the time of the manuscript gain coherence and contextuality when taking alterations into account, especially in the case of deletions. The analysis of alterations in manuscripts, though, is a traditionally very tedious work. In this paper, we present a machine learning-based approach to help categorize alterations in documents. In particular, we present a new probabilistic model (Alteration Latent Dirichlet Allocation, alterLDA in the following) that categorizes content-related alterations. The method proposed here is developed based on experiments carried out on the digital scholarly edition Berlin Intellectuals, for which alterLDA achieves high performance in the recognition of alterations on labelled data. On unlabelled data, applying alterLDA leads to interesting new insights into the alteration behavior of authors, editors and other manuscript contributors, as well as insights into sensitive topics in the correspondence of Berlin intellectuals around 1800.

Weak measurements have been predicted to dramatically alter universal properties of quantum critical wavefunctions, though experimental validation remains an open problem. Here we devise a practical scheme for realizing measurement-altered criticality in a chain of Rydberg atoms tuned to Ising and tricritical Ising phase transitions. In particular, we show that projectively measuring a periodic subset of atoms alters quantum critical correlations in distinct ways that one can control via the choice of measured sites and the measurement outcomes. While our protocol relies on post-selection, the measurement outcomes yielding the most dramatic consequences occur with surprisingly large probabilities: O(10%) with chains featuring O(100) sites. Characterizing the proposed post-measurement states requires only an adjustment in the post-process averaging of outcomes used to characterize unmeasured critical states, resulting in minimal additional experimental overhead for demonstrating measurement-altered criticality.

An increasing number of smart devices controlling loads opens a potential pathway for false data attacks which could alter the loads. The presence of energy storage with its ability to quickly respond to discrepancies in loads offers a promising solution for security by preventing further instabilities and potential blackouts. This paper proposes a control methodology for secure predictive energy management that uses batteries to mitigate the impact of load-altering attacks. To that extent, we develop a microgrid model along with the primary control for microgrid. The developed models and the optimization algorithm are validated through a real-time numerical simulation of a modified IEEE 9 bus system involving a battery as one of the energizing sources. The results show the effectiveness of the battery in countering the load alterations.

Diffusion models have demonstrated exceptional capabilities in generating high-fidelity images. However, their iterative denoising process results in significant computational overhead during inference, limiting their practical deployment in resource-constrained environments. Existing acceleration methods often adopt uniform strategies that fail to capture the temporal variations during diffusion generation, while the commonly adopted sequential pruning-then-fine-tuning strategy suffers from sub-optimality due to the misalignment between pruning decisions made on pretrained weights and the model's final parameters. To address these limitations, we introduce ALTER: All-in-One Layer Pruning and Temporal Expert Routing, a unified framework that transforms diffusion models into a mixture of efficient temporal experts. ALTER achieves a single-stage optimization that unifies layer pruning, expert routing, and model fine-tuning by employing a trainable hypernetwork, which dynamically generates layer pruning decisions and manages timestep routing to specialized, pruned expert sub-networks throughout the ongoing fine-tuning of the UNet. This unified co-optimization strategy enables signific

Is it feasible to alter the ground state properties of a material by engineering its electromagnetic environment? Inspired by theoretical predictions, experimental realizations of such cavity-controlled properties without optical excitation are beginning to emerge. Here, we devised and implemented a novel platform to realize cavity-altered materials. Single crystals of hyperbolic van der Waals (vdW) compounds provide a resonant electromagnetic environment with enhanced density of photonic states and prominent mode confinement. We interfaced hexagonal boron nitride (hBN) with the molecular superconductor $κ$-(BEDT-TTF)$_2$Cu[N(CN)$_2$]Br ($κ$-ET). The frequencies of infrared (IR) hyperbolic modes of hBN match the IR-active carbon-carbon stretching molecular resonance of ($κ$-ET) implicated in superconductivity. Nano-optical data supported by first-principles molecular Langevin dynamics simulations confirm the presence of resonant coupling between the hBN hyperbolic cavity modes and the carbon-carbon stretching mode in ($κ$-ET). Meissner effect measurements via magnetic force microscopy demonstrate a strong suppression of superfluid density near the hBN/($κ$-ET) interface. Non-resona

We investigate how the response of coupled dynamical systems is modified due to a structural alteration of the interaction. The majority of the literature focuses on additive perturbations and symmetrical interaction networks. Here, we consider the challenging problem of multiplicative perturbations and asymmetrical interaction coupling. We introduce a framework to approximate the averaged response at each network node for general structural perturbations, including non-normal and asymmetrical ones. Our findings indicate that both the asymmetry and non-normality of the structural perturbation impact the global and local responses at different orders in time. We propose a set of matrices to identify the nodes whose response is affected the most by the structural alteration.

Despite the significant influx of prompt-tuning techniques for generative vision-language models (VLMs), it remains unclear how sensitive these models are to lexical and semantic alterations in prompts. In this paper, we evaluate the ability of generative VLMs to understand lexical and semantic changes in text using the SugarCrepe++ dataset. We analyze the sensitivity of VLMs to lexical alterations in prompts without corresponding semantic changes. Our findings demonstrate that generative VLMs are highly sensitive to such alterations. Additionally, we show that this vulnerability affects the performance of techniques aimed at achieving consistency in their outputs.

Internal modes of climate variability, such as El Niño and the North Atlantic Oscillation, can have strong influences upon distant weather patterns, effects that are referred to as "teleconnections". The extent to which anthropogenic climate change has and will continue to affect these teleconnections, however, remains uncertain. Here, we employ a covariance fingerprinting approach to demonstrate that shifts in teleconnection patterns affecting monthly temperatures between the periods 1960-1990 and 1990-2020 are attributable to anthropogenic forcing. We further apply multilinear regression to assess the regional contributions and statistical significance of changes in five key climate modes: the El Niño-Southern Oscillation, North Atlantic Oscillation, Southern Annular Mode, Indian Ocean Dipole, and the Pacific Decadal Oscillation. In many regions, observed changes exceed what would be expected from natural variability alone, further implicating an anthropogenic influence. Finally, we provide projections of how these teleconnections will alter in response to further changes in climate.

The successful portrayal of personality in digital characters improves communication and immersion. Current research focuses on expressing personality through modifying animations using heuristic rules or data-driven models. While studies suggest motion style highly influences the apparent personality, the role of appearance can be similarly essential. This work analyzes the influence of movement and appearance on the perceived personality of short videos altered by motion transfer networks. We label the personalities in conference video clips with a user study to determine the samples that best represent the Five-Factor model's high, neutral, and low traits. We alter these videos using the Thin-Plate Spline Motion Model, utilizing the selected samples as the source and driving inputs. We follow five different cases to study the influence of motion and appearance on personality perception. Our comparative study reveals that motion and appearance influence different factors: motion strongly affects perceived extraversion, and appearance helps convey agreeableness and neuroticism.

Cancer is a heterogeneous disease with different combinations of genetic and epigenetic alterations driving the development of cancer in different individuals. While these alterations are believed to converge on genes in key cellular signaling and regulatory pathways, our knowledge of these pathways remains incomplete, making it difficult to identify driver alterations by their recurrence across genes or known pathways. We introduce Combinations of Mutually Exclusive Alterations (CoMEt), an algorithm to identify combinations of alterations de novo, without any prior biological knowledge (e.g. pathways or protein interactions). CoMEt searches for combinations of mutations that exhibit mutual exclusivity, a pattern expected for mutations in pathways. CoMEt has several important feature that distinguish it from existing approaches to analyze mutual exclusivity among alterations. These include: an exact statistical test for mutual exclusivity that is more sensitive in detecting combinations containing rare alterations; simultaneous identification of collections of one or more combinations of mutually exclusive alterations; simultaneous analysis of subtype-specific mutations; and summar

Discontinuous shear thickening (DST) in dense suspensions is accompanied by significant fluctuations in stress at a fixed shear rate. In this work, normal stress fluctuations are shown to have a one-to-one relationship with the formation and dissolution of local high-density regions. Namely, a burst in the force response corresponds to the spontaneous appearance of inhomogeneity. We observe that boundary conditions can significantly alter the spatiotemporal scale of these fluctuations, from short-lived to more sustained and enduring patterns. We estimate the occurrence frequency R and the average intensity Q of individual bursts/inhomogeneity events. The growth of R with the shear rate is the most rapid for the rigid boundary, whereas Q is nonmonotonic with confinement stiffness. Our results indicate that boundary conditions alter the development of inhomogeneity and thus the stress response under shear.

While extensive research has explored the use of large language models (LLMs) for table-based reasoning, most approaches struggle with scalability when applied to large tables. To maintain the superior comprehension abilities of LLMs in these scenarios, we introduce ALTER(Augmentation for Large-Table-Based Reasoning)-a framework designed to harness the latent augmentation potential in both free-form natural language (NL) questions, via the query augmentor, and semi-structured tabular data, through the table augmentor. By utilizing only a small subset of relevant data from the table and supplementing it with pre-augmented schema, semantic, and literal information, ALTER achieves outstanding performance on table-based reasoning benchmarks. We also provide a detailed analysis of large-table scenarios, comparing different methods and various partitioning principles. In these scenarios, our method outperforms all other approaches and exhibits robustness and efficiency against perturbations.

Numerous studies have shown that existing Face Recognition Systems (FRS), including commercial ones, often exhibit biases toward certain ethnicities due to under-represented data. In this work, we explore ethnicity alteration and skin tone modification using synthetic face image generation methods to increase the diversity of datasets. We conduct a detailed analysis by first constructing a balanced face image dataset representing three ethnicities: Asian, Black, and Indian. We then make use of existing Generative Adversarial Network-based (GAN) image-to-image translation and manifold learning models to alter the ethnicity from one to another. A systematic analysis is further conducted to assess the suitability of such datasets for FRS by studying the realistic skin-tone representation using Individual Typology Angle (ITA). Further, we also analyze the quality characteristics using existing Face image quality assessment (FIQA) approaches. We then provide a holistic FRS performance analysis using four different systems. Our findings pave the way for future research works in (i) developing both specific ethnicity and general (any to any) ethnicity alteration models, (ii) expanding suc

Wythoff's Nim is a variant of 2-pile Nim in which players are allowed to take any positive number of stones from pile 1, or any positive number of stones from pile 2, or the same positive number from both piles. The player who makes the last move wins. It is well-known that the P-positions (losing positions) are precisely those where the two piles have sizes $\{\lfloor φn \rfloor, \lfloor φ^2n \rfloor \}$ for some integer $n\geq 0$, and $φ= (1+\sqrt{5})/2 = 1.6180\cdots$. In this paper we consider an altered form of Wythoff's Nim where an arbitrary finite set of positions are designated to be P or N positions. The values of the remaining positions are computed in the normal fashion for the game. We prove that the set of P-positions of the altered game closely resembles that of a translated normal Wythoff game. In fact the fraction of overlap of the sets of P-positions of these two games approaches $1$ as the pile sizes being considered go to infinity.

Fingerprint alteration, also referred to as obfuscation presentation attack, is to intentionally tamper or damage the real friction ridge patterns to avoid identification by an AFIS. This paper proposes a method for detection and localization of fingerprint alterations. Our main contributions are: (i) design and train CNN models on fingerprint images and minutiae-centered local patches in the image to detect and localize regions of fingerprint alterations, and (ii) train a Generative Adversarial Network (GAN) to synthesize altered fingerprints whose characteristics are similar to true altered fingerprints. A successfully trained GAN can alleviate the limited availability of altered fingerprint images for research. A database of 4,815 altered fingerprints from 270 subjects, and an equal number of rolled fingerprint images are used to train and test our models. The proposed approach achieves a True Detection Rate (TDR) of 99.24% at a False Detection Rate (FDR) of 2%, outperforming published results. The synthetically generated altered fingerprint dataset will be open-sourced.

Dirac has predicted that the $g$ factor of an electron is strictly equal to 2 in the framework of relativistic quantum mechanics. However, later physicists have found that this factor can be slightly deviated from 2 (i.e., the problem of anomalous magnetic moments of leptons) when they consider quantum filed theory. This fact thus renders the $g$ factors of free leptons serving as precision tests for quantum electrodynamics, the standard model and beyond. In this work, we re-examine the problem of $g$ factor within the framework of relativistic quantum mechanics. We propose a possible mechanism called the ``electron-braidon mixing'', such that the $g$ factor of an electron can be visibly altered. Our results are hopeful to be verified in experiments and also shed new light to the problem of the anomalous magnetic moments of leptons.

For many people, anxiety, depression, and other social and mental factors can make composing text messages an active challenge. To remedy this problem, large language models (LLMs) may yet prove to be the perfect tool to assist users that would otherwise find texting difficult or stressful. However, despite rapid uptake in LLM usage, considerations for their assistive usage in text message composition have not been explored. A primary concern regarding LLM usage is that poor public sentiment regarding AI introduces the possibility that its usage may harm perceptions of AI-assisted text messages, making usage counter-productive. To (in)validate this possibility, we explore how the belief that a text message did or did not receive AI assistance in composition alters its perceived tone, clarity, and ability to convey intent. In this study, we survey the perceptions of 26 participants on 18 randomly labeled pre-composed text messages. In analyzing the participants' ratings of message tone, clarity, and ability to convey intent, we find that there is no statistically significant evidence that the belief that AI is utilized alters recipient perceptions. This provides hopeful evidence tha

The Grobe--Eberly doublet phenomenon occurs in photoelectron distributions when a field dresses the remaining ion. Its manifestation is due to entanglement between a free electron and a hybrid state of light and matter. Direct detection of such entanglement is however not possible by coincidence schemes due to the dressing mechanism having an inconspicuous phase correlation effect on the ion. Here, it is shown that odd envelopes fundamentally alter the entanglement, such that channel-resolved photoelectron distributions become identifiable in coincidence with the internal state of the field-free ion. This constitutes a first usage of the parity of time symmetry in strong-field interactions.