The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in \emph{misalignment}. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as \emph{realignment}, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.
What is the role of algebra in classical mathematics education? How does it relate to the four quadrivial arts? These questions have troubled the mathematical community since the introduction of algebra into the Renaissance academy by men like François Viète, Guillame Gosselin, and René Descartes. Their challenge is perhaps most starkly articulated at the conclusion of Viète's Introduction to the Analytic Art, where he claims that his algebra "appropriates to itself by right the proud problem of problems, which is: [sic] TO LEAVE NO PROBLEM UNSOLVED". Some contemporary educators respond by eschewing these methods to avoid the excessive formalization often accompanying algebra, and to give a central place to the geometrical tradition of Euclid's Elements. Others embrace the rise of algebra in the curriculum, focusing on contemporary techniques and priorities. This paper seeks to reconcile these perspectives by clarifying the way in which algebra participates in the quadrivial arts. Based on testimony from both the origins of algebra and its contemporary practitioners, I argue that algebra is not so much an eighth liberal art as an arithmetical language of form; an actualized potential in arithmetic. I conclude by offering curricular recommendations which provide glimpses of the practical insights available from this vantage.
Vision encoders are indispensable for allowing impressive performance of Multi-modal Large Language Models (MLLMs) in vision language tasks such as visual question answering and reasoning. However, existing vision encoders focus on global image representations but overlook fine-grained regional analysis. They are limited in fine grained perception due to the scarcity of fine grained annotated data and the lack of a fine grained pre-training paradigm. In this paper, we propose GranViT, a novel Vision Transformer that integrates fine-grained feature extraction with semantic alignment to Large Language Models (LLMs) via region level autoregressive training. We first construct Gran-29M, a dataset comprising 2million natural and OCR images paired with over 180 million high-quality region-level annotations, to enable large scale fine grained pretraining. Consequently, we develop a pretraining-adaptation framework along with a self distillation mechanism to train fine-grained GranViT on Gran-29M. We sufficiently exploit the fine-grained annotations from Gran-29M to resort to bounding-box-to-caption regression to enhance localized visual representation of the vision encoder in the pretraining and caption-to-bounding-box regression to improve vision feature utilization and localization for LLM in the adaptation. We further incorporate a self distillation mechanism that imposes explicit localization constraints on the vision encoder to strengthen its regional reasoning capability. Extensive experiments show that GranViT surpasses existing vision encoders and attains strong transferability to varying LLMs. Remarkably, it achieves state-of-the-art results on fine-grained recognition, multimodal VQA, and OCR understanding.
Fine-tuning is arguably the most straightforward way to tailor a pre-trained model (e.g., a foundation model) to downstream applications, but it also comes with the risk of losing valuable knowledge the model had learned in pre-training. For example, fine-tuning a pre-trained classifier capable of recognizing a large number of classes to master a subset of classes at hand is shown to drastically degrade the model's accuracy in the other classes it had previously learned. As such, it is hard to further use the fine-tuned model when it encounters classes beyond the fine-tuning data. In this paper, we systematically dissect the issue, aiming to answer the fundamental question, "What has been damaged in the fine-tuned model?" To our surprise, we find that the fine-tuned model neither forgets the relationship among the other classes nor degrades the features to recognize these classes. Instead, the fine-tuned model often produces more discriminative features for these other classes, even if they were missing during fine-tuning! {What really hurts the accuracy is the discrepant logit scales between the fine-tuning classes and the other classes}, implying that a simple post-processing calibration would bring back the pre-trained model's capability and at the same time unveil the feature improvement over all classes. We conduct an extensive empirical study to demonstrate the robustness of our findings and provide preliminary explanations underlying them, suggesting new directions for future theoretical analysis. Our code is available at https://github.com/OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated.
Tomando como caso de estudo o projeto NEVE INSULAR, pretendo mostrar como as práticas relacionais de arte e design podem contribuir para a criatividade social e para a sustentabilidade no sentido em que vão ao arrepio da dicotomia entre inovação/tradição. Produto da narrativa histórica ocidental, a dicotomia tradição/inovação equaciona a modernidade ocidental como o telos universal e, concomitantemente, tudo o que resiste à sua conversão – os diferentes saberes ou culturas - como “não-moderno”, “tradicional”. Dentro desta narrativa, o “moderno” é concebido como a aceleração da mudança e o avanço no tempo através do anterior, e o “tradicional” como o estático, fixo no tempo. Neste contexto, a dicotomia tradição/inovação fomenta uma visão atomista do indivíduo, separado da natureza e do contexto cultural, contribuindo para a naturalização das diferenças sociais e para uma visão antropocêntrica do mundo, estando, assim, no cerne da exploração das pessoas e da natureza.
Defendo aqui que a mudança e a inovação, não implicam a ultrapassagem do tradicional pelo moderno mas que, pelo contrário, ambos constituem um fluxo que alimenta o património cultural material e imaterial enquanto parte de um processo político de reinterpretações e de criação de significado histórico.
معصومه سادات نظم, امید سپهری راد, رجبعلی وثوقی مطلق
بررسی جایگاه صنعت پارچهبافی در ایران و تحول و تکامل این هنر، در زمینۀ طرح، نقش و سبک هنری آن، پس از یک دورۀ رکود همزمان با حمله مغول، موردتوجه میباشد. عوامل تأثیرگذار بر این تحول، گسترش و توسعه مراکز و کارگاههای هنری، حضور صنعتگران خلاق از یکسو و سیاستهای حاکمان از سوی دیگر، میباشد؛ بنابراین عوامل ایجاد تحول در صنعت پارچهبافی عصر ایلخانی و تیموری بسیار مهم و دارای اهمیت است. این مقاله بر آن است با استفاده از روش تحقیق تاریخی و بهصورت توصیفی و تحلیلی انجام شده است. سؤال پژوهش این است که سیاستهای حاکمان چه تأثیری در تکامل صنعت پارچهبافی عصر ایلخانی و تیموری داشته و چه تحولی در طرح و نقش و سبک هنری ایجاد کرده است؟ از نتایج بهدستآمده این پژوهش میتوان به تکامل و تحول در صنعت پارچهبافی بهویژه در زمینههای تولید انواع پارچه، طرح و نقش و تکنیکهای متعدد، برجستهکاری، مصورسازی پارچه در مراکز و کارگاههای پارچهبافی تحت حمایت حاکمان، و در نهایت پدیدآمدن سبک جدید هنری تلفیقی سنتی ایرانی و چینی میباشد، اشاره کرد.اهداف پژوهش:بررسی تأثیر سیاستهای حکام ایلخانی و تیموری بر پارچهبافی.بررسی تحولات پارچهبافی ایلخانی و تیموری از نظر و نقش و سبک هنری.سؤالات پژوهش:سیاستهای حکام ایلخانی و تیموری چه تأثیری بر صنعت پارچهبافی داشت؟در دورۀ ایلخانی و تیموری چه تحولی در نقش و سبک هنری پارچهها ایجاد شد؟
This paper presents an analytical framework for evaluating the impact of the two main processes identified in Madrid timber-framed courtyard buildings reinforcing the courtyard's crucial role in architectural heritage sustainability: their development caused dwellings to darken (1737–1950), and the existing challenge is to open new patios according to the listed buildings' protection policy.Critical findings in archival research and existing building assessments led to completed data collection and analysis. When exploring construction expansion, from the house towards the collective prototype around a gradually reduced patio and dwelling darkening (1737–1786), buildings resulted in open, semiopen, underlighted and closed designs. Study area on-site assessment uncovered the transition from the older low-rise type (1–3) floors to the modern closed model of up to (4–6) storeys.Construction enlargements, consequently reducing original green areas and courtyards, call for reversing that process or formulating a new strategy. It debates the partial demolition of listed buildings in upgrades and its agreement with the needed “blue courtyard” strategy (PGOUM), resulting from joining new rear individual patios to create a communal building block courtyard. The uncovered “in-between” type (1864), which integrates two open-air collective spaces, an inner courtyard and a back garden, appears challenging to explore.
This paper looks at the polyphonic analysis of Moroccan caricature in the context of the Covid-19 pandemic. It highlights the importance of the image as an information tool within an area of multiple communication. As a form of visual expression, caricature aims to convey information concisely and persuasively, while reflecting the perspective of its creator. The present study sets out to understand caricature by adopting a linguistic approach and focusing specifically on the analysis of linguistic polyphony in Moroccan caricatures related to the global health crisis. This approach adopts Ducrot's enunciative theory as an analytical framework. Based on Imad Sanouni's caricatures addressing this theme, the present analysis highlights the diversity of voices and perspectives within a caricatural image image, faithfully illustrating the crisis experienced by Moroccan society during the pandemic. Results show that these drawings faithfully reflect this epidemiological period through the means of iconographic and linguistic codes to create layers of meaning and discourse. By combining visual and linguistic elements, this study demonstrates that caricature could constitute a polyphonic means of visual communication, contributing
Résumé : Cet article présente une étude scientifique sur la linguistique et la didactique. Les deux domaines sont considérés comme un pilier essentiel en classe du français langue étrangère. La présente étude concerne précisément la compréhension orale, étant une compétence primordiale de la communication orale, car l'apprenant est naturellement exposé à l'écoute de la langue avant de la produire. Sur la base de cette logique, mener une enquête auprès d'un groupe d'élèves du secondaire qualifiant marocain en vue d’identifier les erreurs liées à une telle compétence et leur origine, reste la première étape didactique pour concevoir des dispositifs pédagogiques dans les futurs projets de recherche sur le même sujet. Cette recherche porte également sur l'analyse de certaines erreurs de compréhension orale recueillies à partir de l'enquête déjà citée ; cette analyse est basée sur des cadres contextuel, épistémologique et théorique qui régissent la compréhension orale et l'analyse de l'erreur.
Mots-clés : didactique, FLE, erreur, compétence grammaticale, compréhension orale.
Arts in general, Computational linguistics. Natural language processing
Visually complex texts emerge as a response to complicated cosmological, social, or ontological phenomena in reality, and different pre-modern cultures came up with their own unique ways of re-presenting this reality in textual form. Today, with the help of digital media, it becomes possible to discuss visually complex texts from different cultures in a way that facilitates the exchange of research ideas between specialists working with different regions and languages.
Rileggendo gli scritti di Joris Karl Huysmans, autore poliedrico e assai sensibile al mondo dell’arte, l’articolo indaga i suoi interventi, numerosi quantitativamente e specifici qualitativamente, dedicati ai “primitivi” fiamminghi. Il tema dei “primitivi”, sia riguardo al mondo nordico sia in senso generale, è da tempo oggetto di ampio interesse storiografico. Ma Huysmans dimostra un’attenzione ed una qualità specifica assai precoce, in anticipo sul grande dibattito novecentesco, che esula da una generica rivisitazione letteraria, presente invece in altri autori e che fa di lui una moderna figura di critico d’arte, attento a problemi stilistici, attributivi e qualitativi.
The paper focuses on Joris Karl Huysmans, the famous author and art critic and as is well know, passionate about Moreau and Redon. In addition to this interest, many pages of his novels are dedicated to a very interesting topic: the Flamish Primitives Painters. From a general point of view, their fortune over the centuries has been established from quite a long time. But Huysmans, unlike most writers, exhibits a specific competence and can be considered as a true art critic, focused on matter of style, quality and connoisseurship.
Given a bounded finely open set $V$ and a function $f$ on the fine boundary of $V$, we introduce four types of upper Perron solutions to the nonlinear Dirichlet problem for $p$-energy minimizers, $1<p<\infty$, with $f$ as boundary data. These solutions are given as pointwise infima of suitable families of fine $p$-superminimizers in $V$. We show (under natural assumptions) that the four upper Perron solutions are equal quasieverywhere and that they are fine $p$-minimizers of the $p$-energy integral. We moreover show that the upper and lower Perron solutions coincide quasieverywhere for Sobolev and for uniformly continuous boundary data, i.e.\ that such boundary data are resolutive. For the uniformly continuous boundary data, the Perron solutions are also shown to be finely continuous and thus finely $p$-harmonic. We prove our results in a complete metric space $X$ equipped with a doubling measure supporting a $p$-Poincaré inequality, but they are new also in unweighted $\mathbf{R}^n$.
Andrei E. Tarkhov, A. V. Rozhkov, Alfred Zong
et al.
Charge-density-wave order in a solid can be temporarily "melted" by a strong laser pulse. Here we use the discrete Gross-Pitaevskii equation on a cubic lattice to simulate the recovery of the CDW long-range phase coherence following such a pulse. Our simulations indicate that the recovery process is dramatically slowed down by the three-dimensional topological defects - CDW dislocations - created as a result of strongly nonequilibrium heating and cooling of the system. Overall, the simulated CDW recovery was found to be remarkably reminiscent of a recent pump-probe experiment in LaTe$_3$.
Recently Bowden, Hensel and Webb defined the fine curve graph for surfaces, extending the notion of curve graphs for the study of homeomorphism or diffeomorphism groups of surfaces. Later Long, Margalit, Pham, Verberne and Yao proved that for a closed surface of genus $g\geqslant 2$, the automorphism group of the fine graph is naturally isomorphic to the homeomorphism group of the surface. We extend this result to the torus case $g=1$; in fact our method works for more general surfaces, compact or not, orientable or not. We also discuss the case of a smooth version of the fine graph.
Arts are a seamless way to introduce the general public to both basic and more sophisticated astronomical concepts. The visual richness of astronomy makes it attractive and easily incorporated in painting and literature. Astronomy is the only science with a muse - Urania - implying that, at least in the eyes of the ancients, it was an art itself. I review some less well known representation of astronomical concepts in literature with potential application in education.
The unrestrained and constant inflow of tourists to some cities can provoke
the discomfort of residents. The term ‘overtourism’ describes a condition in
which the limits are exceeded to the point that regular everyday urban life
suffers. The quantity of visitors and resulting crowds create burdens on the
functioning of some city services, and they simultaneously result in a
deterioration in the quality of sightseeing. This paper gives a review of
the academic literature concerning this issue, especially studies
concentrating on cruising and alternative short-term renting. Major
complaints raised by locals on the negative impact of tourism are not only
about overcrowding, but also include the serious social and economic
questions of the endangerment of rights. In order to regulate and mitigate
conflicts, preserve identity and allow decent urban life in all its aspects,
city governments are forced to introduce some limitations and rules in the
sphere of tourism, and also with regard to the real-estate market and urban
planning. The level of measures and policies required depend on the type of
problem, size of the city and dispersion of the attractions.
Iacopo Bertelli, Joris J. Carmiggelt, Tao Yu
et al.
Spin waves - the elementary excitations of magnetic materials - are prime candidate signal carriers for low dissipation information processing. Being able to image coherent spin-wave transport is crucial for developing interference-based spin-wave devices. We introduce a platform for probing coherent spin waves based on magnetic resonance imaging with electron spins in diamond. Focusing on a thin-film magnetic insulator, we quantify spin-wave amplitudes, visualize the dispersion, and demonstrate time-domain measurements of spin-wave packets. We use our platform to study spin-wave interference, revealing uni-directional, autofocused spin-wave patterns with frequency-controlled numerical apertures. A theoretical analysis explains the patterns in terms of chiral spin-wave excitation and stray-field coupling to the sensor spins. These results pave the way for probing spin waves in atomically thin magnets, even when embedded between opaque materials.
Ali-Mohammad Kamali, Mohammad Taghi Najafi, Mohammad Nami
The reciprocal impact of applied neuroscience and cognitive studies on humanities has been extensive and growing over the past 30 years of research. Studies on neuroaesthetics have provided novel insights in visual arts, music as well as abstract and dramatic art. Neuro-Art is an experimental concept in applied neuroscience where scientists can study the mechanistic pathways involved for instance in visual art through which creativity and artistic capacity might receive further empowerment. Based on the existing evidence, at least 3 large-scale brain networks are involved simultaneously when one is submitted to a creativity-related task. The question whether the key brain regions involved in visual art creativity can be identified and receive neuromodulation to get empowered prompted us to perform the present case investigation. Virtual reality and functional quantitative electroencephalography upon 2- vs 3-dimentional painting were employed to study cortical neurodynamics in a professional painting artist.