Hasil untuk "Speculative philosophy"

Qitan Lv, Tianyu Liu, Wen Wu et al.

Speculative decoding (SD) has emerged as a promising approach to accelerate LLM inference without sacrificing output quality. Existing SD methods tailored for video-LLMs primarily focus on pruning redundant visual tokens to mitigate the computational burden of massive visual inputs. However, existing methods do not achieve inference acceleration comparable to text-only LLMs. We observe from extensive experiments that this phenomenon mainly stems from two limitations: (i) their pruning strategies inadequately preserve visual semantic tokens, degrading draft quality and acceptance rates; (ii) even with aggressive pruning (e.g., 90% visual tokens removed), the draft model's remaining inference cost limits overall speedup. To address these limitations, we propose HIPPO, a general holistic-aware parallel speculative decoding framework. Specifically, HIPPO proposes (i) a semantic-aware token preservation method, which fuses global attention scores with local visual semantics to retain semantic information at high pruning ratios; (ii) a video parallel SD algorithm that decouples and overlaps draft generation and target verification phases. Experiments on four video-LLMs across six benchmarks demonstrate HIPPO's effectiveness, yielding up to 3.51x speedup compared to vanilla auto-regressive decoding.

Anqi Wang, Yue Hua, Xinyue Zhang et al.

Hyper-learning and Unlearning is a speculative animation that reflect how learning is reconfigured within digital media ecologies. Using architectural education as a microcosm, the work reframes the city as a hyper-learning apparatus where urban space, algorithmic systems, and platform infrastructures condition cognition and agency. By staging both hyper-learning and the unlearning induced by machine-supported cognition, the work critiques institutional gatekeeping while revealing how platforms reshape expertise, memory, and spatial experience. This project invites viewers to reconsider how urban space becomes pedagogical infrastructure in a posthumanism era.

Satyam Goyal, Kushal Patel, Tanush Mittal et al.

Masked Diffusion Models (MDMs) offer a promising alternative to autoregressive language models by enabling parallel token generation and bidirectional context modeling. However, their inference speed is significantly limited by the inability to cache key-value pairs due to bidirectional attention, requiring $O(N^2)$ computations at each generation step. While recent methods like FastDLLM and DkvCache improve inference speed through attention approximations and caching strategies, they achieve speedups at the cost of generation quality. We propose DualDiffusion, a speculative decoding framework for MDMs that combines fast drafter models (using efficient approximations) with slower, more accurate verifier models. By running multiple steps of a lightweight drafter followed by a single verification step, DualDiffusion achieves a superior Pareto frontier between generation steps and accuracy compared to existing approaches. We evaluate our method on MMLU and GSM8K, demonstrating that DualDiffusion maintains high accuracy while reducing the number of generation steps required, effectively pushing the quality-efficiency trade-off curve for masked diffusion language models.

I. V. Kocheshkova, S. A. Rumbest, A. Ya. Mindlina

Relevance. Currently, the incidence of vaccine-controlled infections tends to increase. Despite the fact that some of them are not registered at all, the situation remains unstable and tense. Vaccination is the most effective preventive measure. The effectiveness of immunization mainly depends on achieving the necessary vaccination coverage and meeting the deadlines for their implementation. Aim. Assess the coverage of vaccination and revaccination against diphtheria, pertussis, measles, rubella and mumps in the period from 2019 to 2022. on the territory of the Russian Federation, as well as to analyze these indicators for 2022, including the timeliness of immunization within the framework of the National Calendar of Preventive Vaccinations.Materials and methods. A retrospective epidemiological analysis of the state of vaccination coverage, revaccination and the timeliness of their implementation was carried out. Data from the Federal statistical observation form No. 6 «Information on the populations of children and adults vaccinated against infectious diseases» for 2019–2022 were used.Results. Vaccination coverage against diphtheria and pertussis at the age of 0–1 years in the Russian Federation from 2020 is only 47.31% and 45.97% respectively. Of these, 89.31% were vaccinated against diphtheria at the age of 6 months - 11 months, 29 days, and 88,65% against whooping cough. The timeliness of vaccination against these infections is achieved only by 12 months: about 97% have been vaccinated by this age. The average levels of vaccination coverage against measles, rubella and mumps in 1 years – 1 years, 11 months, 29 days were 96,47%, 96,42% and 96,46%, respectively. The rate of timely immunization reaches 97% only by 24 months. RV1 coverage in the age group of 1 years – 1 years, 11 months, 29 days is 50.87% against diphtheria and 50.67% against pertussis, in the group of 2 years – 2 years, 11 months, 29 days – 96,38% and 96,17%, respectively. 96% were vaccinated in a timely manner by 24 months. Coverage of the second revaccination against diphtheria at the age of 6 years – 6 years, 11 months, 29 days, is on average 54.37%, in the age group of 7 years – 7 years, 11 months, 29 days – 95.88%. Coverage of RV3 at the age of 14 years – 14 years, 11 months, 29 days – 95.53%, and RV4 in the age groups of 18 years – 35 years, 11 months, 29 days, 36 years – 59 years, 11 months, 29 days, and 60 years and older in 2022, amounted to 71.2%, 78.3% and 73.9%, respectively. The coverage of revaccination against measles, rubella and mumps in 6 years – 6 years,11 months, 29 days, amounted to 96,06%, 96,02% and 96,04% respectively.Conclusion. The analysis showed that vaccination is carried out much later than the deadlines set by the National Calendar of Preventive Vaccinations. Vaccination and revaccination coverage against diphtheria and pertussis, measles, rubella and mumps does not reach the required level of 95% in all regions, which poses a threat to the epidemiological well-being of the population. It should be noted that measles vaccination coverage does not reach 98%, thus conditions for its elimination are not created. In 2020, diphtheria, whooping cough, measles, rubella and mumps saw a decrease in vaccination and revaccination coverage due to the epidemic of COVID-19, but by 2022 coverage still did not reach the values of 2019.

Zijian Lin, Yang Zhang, Yougen Yuan et al.

Modern autoregressive speech synthesis models leveraging language models have demonstrated remarkable performance. However, the sequential nature of next token prediction in these models leads to significant latency, hindering their deployment in scenarios where inference speed is critical. In this work, we propose Speech Speculative Decoding (SSD), a novel framework for autoregressive speech synthesis acceleration. Specifically, our method employs a lightweight draft model to generate candidate token sequences, which are subsequently verified in parallel by the target model using the proposed SSD framework. Experimental results demonstrate that SSD achieves a significant speedup of 1.4x compared with conventional autoregressive decoding, while maintaining high fidelity and naturalness. Subjective evaluations further validate the effectiveness of SSD in preserving the perceptual quality of the target model while accelerating inference.

Hossein Entezari Zarch, Lei Gao, Chaoyi Jiang et al.

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.

Yichao Fu, Rui Ge, Zelei Shao et al.

Reasoning models excel by generating long chain-of-thoughts, but decoding the resulting thousands of tokens is slow. Token-level speculative decoding (SD) helps, but its benefit is capped, because the chance that an entire $γ$-token guess is correct falls exponentially as $γ$ grows. This means allocating more compute for longer token drafts faces an algorithmic ceiling -- making the speedup modest and hardware-agnostic. We raise this ceiling with Lookahead Reasoning, which exploits a second, step-level layer of parallelism. Our key insight is that reasoning models generate step-by-step, and each step needs only to be semantically correct, not exact token matching. In Lookahead Reasoning, a lightweight draft model proposes several future steps; the target model expands each proposal in one batched pass, and a verifier keeps semantically correct steps while letting the target regenerate any that fail. Token-level SD still operates within each reasoning step, so the two layers of parallelism multiply. We show Lookahead Reasoning lifts the peak speedup of SD both theoretically and empirically. Across GSM8K, AIME, and other benchmarks, Lookahead Reasoning improves the speedup of SD from 1.4x to 2.1x while preserving answer quality, and its speedup scales better with additional GPU throughput. Our code is available at https://github.com/hao-ai-lab/LookaheadReasoning

Tin Trung Nguyen, Jiannan Xu, Zora Che et al.

Although popularized AI fairness metrics, e.g., demographic parity, have uncovered bias in AI-assisted decision-making outcomes, they do not consider how much effort one has spent to get to where one is today in the input feature space. However, the notion of effort is important in how Philosophy and humans understand fairness. We propose a philosophy-informed approach to conceptualize and evaluate Effort-aware Fairness (EaF), grounded in the concept of Force, which represents the temporal trajectory of predictive features coupled with inertia. Besides theoretical formulation, our empirical contributions include: (1) a pre-registered human subjects experiment, which shows that for both stages of the (individual) fairness evaluation process, people consider the temporal trajectory of a predictive feature more than its aggregate value; (2) pipelines to compute Effort-aware Individual/Group Fairness in the criminal justice and personal finance contexts. Our work may enable AI model auditors to uncover and potentially correct unfair decisions against individuals who have spent significant efforts to improve but are still stuck with systemic disadvantages outside their control.

Feiye Huo, Jianchao Tan, Kefeng Zhang et al.

The growing scale of Large Language Models (LLMs) has exacerbated inference latency and computational costs. Speculative decoding methods, which aim to mitigate these issues, often face inefficiencies in the construction of token trees and the verification of candidate tokens. Existing strategies, including chain mode, static tree, and dynamic tree approaches, have limitations in accurately preparing candidate token trees for verification. We propose a novel method named C2T that adopts a lightweight classifier to generate and prune token trees dynamically. Our classifier considers additional feature variables beyond the commonly used joint probability to predict the confidence score for each draft token to determine whether it is the candidate token for verification. This method outperforms state-of-the-art (SOTA) methods such as EAGLE-2 on multiple benchmarks, by reducing the total number of candidate tokens by 25% while maintaining or even improving the acceptance length.

Yilong Zhao, Jiaming Tang, Kan Zhu et al.

Reasoning language models have demonstrated remarkable capabilities on challenging tasks by generating elaborate chain-of-thought (CoT) solutions. However, such lengthy generation shifts the inference bottleneck from compute-bound to memory-bound. To generate each token, the model applies full attention to all previously generated tokens, requiring memory access to an increasingly large KV-Cache. Consequently, longer generations demand more memory access for every step, leading to substantial pressure on memory bandwidth. To address this, we introduce SparseSpec, a speculative decoding framework that reuses the same model as the draft and target models (i.e., self-speculation). SparseSpec features a novel sparse attention mechanism, PillarAttn, as the draft model, which accurately selects critical tokens via elegantly reusing information from the verification stage. Furthermore, SparseSpec co-designs self-speculation with three system innovations: (1) a unified scheduler to batch token drafting and verification, (2) delayed verification for CPU/GPU overlap, and (3) dynamic KV-Cache management to maximize memory utilization. Across various models and datasets, SparseSpec outperforms state-of-the-art solutions, with an up to 2.13x throughput speedup.

Elena Grishina

The article considers the type of philosophical constructions of B. P. Vysheslavtsev from the point of view of the correlation between the categories of ideal and real. Agreeing with Plotinus and putting forward the three-stage ontological model as a universal and the most complete one (in contrast to the transcendentalist and positivist ones), he unites all forms of human activity into the cultural activity-creative cosmos, within which, in his opinion, the synthesis of the ideal and the real takes place, as well as the realisation of the highest spiritual goal - cognition and experience of the transcendent Absolute. In the context of foreign and domestic philosophical discussions of the first half of the 20th century, Fichte-s practical philosophy becomes relevant, which lays down a new type of idealism centred on the real as the realm of empirical experience in its concrete unitary forms. Spiritual being (the totality of embodied meanings, goals and values), which is based not on abstract principles, but on the dynamics of absolute-like self, its continuous transcendence, is a harmonious conjugation of opposite beginnings, and thanks to this philosophy goes beyond the limits of philosophical discourse proper and becomes the philosophy of life, existentialism. The article proposes the division of ideal-realist philosophy into two types. Religious-philosophical realism (ideal-realism with a three-stage ontological model), to which B. P. Vysheslavtsev adheres, forms a new idea of reality as «belonging to the Absolute». The theme of the «middle» is one of the ideal-realistic motifs. This is the theme of combining two original, i.e. ontologically different and different quality (and, therefore, antinomic) principles - the ideal and the real, and the philosophical systems associated with these principles (dogmatic idealism and dogmatic idealism) into a kind of dualistic unity pointing to the Absolute, its transcendental-immanent position and special meta-manifestation in the world. Accessible to rational knowledge and formal-logical speculative analysis in the form of the law of the unity of opposites, B. P. Vysheslavtsev insists that the very source of this law has neither unity nor opposites, but is the center of divine secrets, the living harmonious self, which «holds» this unity.

Yunfei Cheng, Aonan Zhang, Xuanyu Zhang et al.

We present Recurrent Drafter (ReDrafter), an advanced speculative decoding approach that achieves state-of-the-art speedup for large language models (LLMs) inference. The performance gains are driven by three key aspects: (1) leveraging a recurrent neural network (RNN) as the draft model conditioning on LLM's hidden states, (2) applying a dynamic tree attention algorithm over beam search results to eliminate duplicated prefixes in candidate sequences, and (3) training through knowledge distillation from the LLM. ReDrafter accelerates Vicuna inference in MT-Bench by up to 2.8x with a PyTorch implementation on Nvidia H100 GPUs. To demonstrate its practicality in real environments, we also validated its effectiveness for on-device applications by implementing the approach in MLX and benchmarking performance on Metal GPUs in Apple Silicon chips, achieving up to 2.3x speedup.

Abdoul Jalil Djiberou Mahamadou, Aloysius Ochasi, Russ B. Altman

Data are essential in developing healthcare artificial intelligence (AI) systems. However, patient data collection, access, and use raise ethical concerns, including informed consent, data bias, data protection and privacy, data ownership, and benefit sharing. Various ethical frameworks have been proposed to ensure the ethical use of healthcare data and AI, however, these frameworks often align with Western cultural values, social norms, and institutional contexts emphasizing individual autonomy and well-being. Ethical guidelines must reflect political and cultural settings to account for cultural diversity, inclusivity, and historical factors such as colonialism. Thus, this paper discusses healthcare data ethics in the AI era in Africa from the Ubuntu philosophy perspective. It focuses on the contrast between individualistic and communitarian approaches to data ethics. The proposed framework could inform stakeholders, including AI developers, healthcare providers, the public, and policy-makers about healthcare data ethical usage in AI in Africa.

Xiangguo Liu, Ruochen Jiao, Yixuan Wang et al.

Recently significant progress has been made in vehicle prediction and planning algorithms for autonomous driving. However, it remains quite challenging for an autonomous vehicle to plan its trajectory in complex scenarios when it is difficult to accurately predict its surrounding vehicles' behaviors and trajectories. In this work, to maximize performance while ensuring safety, we propose a novel speculative planning framework based on a prediction-planning interface that quantifies both the behavior-level and trajectory-level uncertainties of surrounding vehicles. Our framework leverages recent prediction algorithms that can provide one or more possible behaviors and trajectories of the surrounding vehicles with probability estimation. It adapts those predictions based on the latest system states and traffic environment, and conducts planning to maximize the expected reward of the ego vehicle by considering the probabilistic predictions of all scenarios and ensure system safety by ruling out actions that may be unsafe in worst case. We demonstrate the effectiveness of our approach in improving system performance and ensuring system safety over other baseline methods, via extensive simulations in SUMO on a challenging multi-lane highway lane-changing case study.

Yves Roullière

What Emmanuel Falque seems to achieve in this book is to make the "out of phenomenon" synonymous with the "tragic". Certainly, we see the heuristic interest in creating and cultivating the concept of "out of phenomenon". In Aeschylus' Agamemnon (a reference that runs throughout this book), the atrocious rubs shoulders with the filthy and can only test us, traumatize us, and, therefore, can only "modify" us in some way. If it is also true that the tragic, according to Kierkegaard in particular, can lead to despair, another attitude seems to be dodged and no less sketched by Falque, namely, under the influence of Nietzsche this time, the amor fati: "Learned in the War School of life: what does not kill me strengthens me." Falque does not refer to it explicitly, while the approach that consists in looking the trauma in the face, in having no fear of fixing it, of penetrating it, of inhabiting it, appears to stem from this love of fatum, of destiny "that falls on us", enough in any case for us to be modified in a sense as unexpected as irreparable. But isn't it also the definition and very role of tragic and/or catastrophic feelings?

Heitor Nelson Ferreira

Neste trabalho aborda-se a questão que está nos escritos kantianos que propõem um único e mesmo fim da humanidade como a totalidade dos homens unidos sob o ideal de uma sociedade cosmopolita, como na Ideia de uma história universal de um ponto de vista cosmopolita, e numa comunidade ética como n´A religião nos simples limites da Razão. A moralidade é pressuposta como fundamento para realização do fim humano em ambos os textos. Ora, em Kant agir com interesse a fins não está dentro dos requisitos da moralidade, que deve proceder com o princípio da incondicionalidade sem referir-se a propósitos. Logo, esboça-se aqui, grosso modo, uma possível interpretação de como é possível conciliar a proposta de um fim para a humanidade com a ideia de que a autonomia não deve orientar-se por fins.

Israel Hordecte

O objetivo deste trabalho é analisar a noção de vontade de verdade sob o prisma de exercício do poder, a partir das considerações desenvolvidas por Friedrich Nietzsche (1844-1900) e Michel Foucault (1926-1984). Neste contexto, a problemática diz respeito aos modos como a verdade se relaciona com o humano, tanto para compreender a existência, como denuncia Nietzsche, quanto para produções de discurso, como indica Foucault. Assim, buscar-se-á responder: “A vontade de verdade restringe a capacidade humana de interpretar a existência? E, ainda: quais os limites estabelecidos entre a verdade e o humano, para que este continue atuando sob a perspectiva da superação nietzschiana e da subjetivação foucaultiana?”. Com isso em vista, serão utilizadas as obras A Gaia Ciência (1882) e Para a Genealogia da Moral (1887), de Nietzsche, em que o autor aborda a vontade de verdade a partir do modo como esta se desenvolve em paralelo à filosofia socrático-platônica, além dos reflexos desta no cristianismo, que se configura em vontade de domínio no ideal ascético. Não obstante, será analisado o posicionamento de Foucault em A Ordem do Discurso (1971), interpretando a vontade de verdade enquanto regra do discurso que promove uma forma de exercício do poder dentro da sociedade e impede, por sua vez, a subjetivação do sujeito através do dizer-verdadeiro. Desse modo, será possível entrever, ainda, uma relação teórica que tange os modos como Nietzsche e Foucault avaliam a noção de vontade de verdade e os seus desdobramentos quando associados à figura humana em sociedade.

Carlos Estellita-Lins

Definitely dated, ephemeral and circumstantial, the paper considers the concept of event (événement) during the pandemic of the new covid-19 SARS-2 in the world, listening from the cloister experience springing out of quarantine in a small apartment in Copacabana. The recent and sudden need to describe illness experience by philosophers, sociologists, anthropologists and intellectuals has been envisaged as a paradox: the search for solutions and the perception of their impossibility emerge simultaneously. The status of today's philosophical reflections concerning the end of all things is mentioned through an essayistic and non-exhaustive way. The issue of anthropocene and zoonosis is addressed in this context of change, where modes, an insurmountable etymology of the modern project, are confronted with eschatological impasses and altogether with the postponement of the end of the world.

Minh N. H. Nguyen, Shashi Raj Pandey, Kyi Thar et al.

Due to the availability of huge amounts of data and processing abilities, current artificial intelligence (AI) systems are effective in solving complex tasks. However, despite the success of AI in different areas, the problem of designing AI systems that can truly mimic human cognitive capabilities such as artificial general intelligence, remains largely open. Consequently, many emerging cross-device AI applications will require a transition from traditional centralized learning systems towards large-scale distributed AI systems that can collaboratively perform multiple complex learning tasks. In this paper, we propose a novel design philosophy called democratized learning (Dem-AI) whose goal is to build large-scale distributed learning systems that rely on the self-organization of distributed learning agents that are well-connected, but limited in learning capabilities. Correspondingly, inspired by the societal groups of humans, the specialized groups of learning agents in the proposed Dem-AI system are self-organized in a hierarchical structure to collectively perform learning tasks more efficiently. As such, the Dem-AI learning system can evolve and regulate itself based on the underlying duality of two processes which we call specialized and generalized processes. In this regard, we present a reference design as a guideline to realize future Dem-AI systems, inspired by various interdisciplinary fields. Accordingly, we introduce four underlying mechanisms in the design such as plasticity-stability transition mechanism, self-organizing hierarchical structuring, specialized learning, and generalization. Finally, we establish possible extensions and new challenges for the existing learning approaches to provide better scalable, flexible, and more powerful learning systems with the new setting of Dem-AI.

Michiel Min, Chris W. Ormel, Katy Chubb et al.

Aims: ARCiS, a novel code for the analysis of exoplanet transmission and emission spectra is presented. The aim of the modelling framework is to provide a tool able to link observations to physical models of exoplanet atmospheres. Methods: The modelling philosophy chosen in this paper is to use physical and chemical models to constrain certain parameters while keeping free the parts where our physical understanding is still more limited. This approach, in between full physical modelling and full parameterisation, allows us to use the processes we understand well and parameterise those less understood. A Bayesian retrieval framework is implemented and applied to the transit spectra of a set of 10 hot Jupiters. The code contains chemistry and cloud formation and has the option for self consistent temperature structure computations. Results: The code presented is fast and flexible enough to be used for retrieval and for target list simulations for e.g. JWST or the ESA Ariel missions. We present results for the retrieval of elemental abundance ratios using the physical retrieval framework and compare this to results obtained using a parameterised retrieval setup. Conclusions: We conclude that for most of the targets considered the current dataset is not constraining enough to reliably pin down the elemental abundance ratios. We find no significant correlations between different physical parameters. We confirm that planets in our sample with a strong slope in the optical transmission spectrum are the planets where we find cloud formation to be most active. Finally, we conclude that with ARCiS we have a computationally efficient tool to analyse exoplanet observations in the context of physical and chemical models.