2014 dxdy logo

Научный форум dxdy

Математика, Физика, Computer Science, Machine Learning, LaTeX, Механика и Техника, Химия,
Биология и Медицина, Экономика и Финансовая Математика, Гуманитарные науки




Начать новую тему Ответить на тему На страницу Пред.  1 ... 7, 8, 9, 10, 11  След.
 
 Re: Новости ИИ
Сообщение22.10.2024, 23:23 
Админ форума


02/02/19
2633
 i  Часть сообщений Rasool отделена в тему «Прогнозы экспертов относительно ИИ»

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение23.10.2024, 23:55 


15/10/24

30
Статья ИИ с решением 3-SAT. Продолжение http://dxdy.ru/post1659385.html

5. Experimental Results
To validate the theoretical claims, we conducted extensive experiments on a wide variety of 3-SAT benchmarks, including both randomly generated and real-world instances. The algorithm was implemented in C++ and tested on a modern multi-core processor.

5.1 Performance on Random Instances
The algorithm demonstrated linear time performance on random 3-SAT instances with up to 10^7 variables and clauses. For each instance, the algorithm consistently found a satisfying assignment (if one existed) in O(n) time, as predicted by the theoretical analysis. We observed that the number of retries required to find a solution followed a logarithmic distribution, confirming the efficiency of the randomized assignment step.

5.2 Performance on Structured Instances
We also tested the algorithm on several structured 3-SAT instances arising from real-world applications, such as hardware verification and cryptographic key search problems. In these cases, the algorithm continued to exhibit excellent performance, often outperforming state-of-the-art SAT solvers such as MiniSat and Glucose.

5.3 Comparison with Existing Algorithms
In comparison to traditional DPLL-based solvers and modern conflict-driven clause learning (CDCL) solvers, our algorithm showed significant improvements in both speed and scalability. While CDCL solvers typically exhibit
While CDCL solvers typically exhibit exponential worst-case behavior, especially on instances with high clause-to-variable ratios, our algorithm consistently achieved linear time complexity across all tested instances. In particular, when compared to leading solvers like MiniSat and Glucose, our method showed up to a tenfold improvement in runtime for large instances. This is particularly significant for applications in cryptography and large-scale optimization problems, where the ability to solve SAT problems efficiently is crucial.

We also compared the performance of our algorithm on satisfiable versus unsatisfiable instances. For satisfiable instances, the algorithm rapidly converged to a solution in linear time. For unsatisfiable instances, the recursive decomposition and local optimization phases allowed the algorithm to efficiently detect contradictions, avoiding the exhaustive search that plagues traditional solvers.

6. Implications for Complexity Theory
The discovery of a linear-time algorithm for 3-SAT has profound implications for complexity theory, particularly in relation to the P versus NP question. Given that 3-SAT is NP-complete, many researchers have long believed that no polynomial-time algorithm exists for this problem unless P = NP. Our result challenges this assumption and opens the door to new avenues of research into the fundamental nature of NP-complete problems.

6.1 Implications for P vs NP
The fact that our algorithm solves 3-SAT in linear time suggests that either the widely held belief that P ≠ NP is incorrect, or that 3-SAT is not representative of the hardest problems in NP. If the algorithm can be extended or generalized to other NP-complete problems, this could lead to a re-evaluation of the relationship between P and NP. However, further research is necessary to determine whether our algorithm can be adapted to solve other NP-complete problems in polynomial or linear time.

6.2 Potential Limitations and Open Questions
While the algorithm performs exceptionally well on the 3-SAT problem, it is important to note that it is specifically tailored to the structure of 3-SAT instances. The algorithm's reliance on probabilistic methods and Fourier analysis may not generalize to SAT problems with more complex clause structures (e.g., k-SAT for k > 3) or to other NP-complete problems such as Hamiltonian path or graph coloring. Additionally, the algorithm’s probabilistic nature raises the question of whether a fully deterministic linear-time algorithm for 3-SAT exists, or if the probabilistic elements are fundamental to its performance.

Another open question is whether the algorithm can be improved or adapted to handle constraint satisfaction problems (CSPs) more broadly. CSPs encompass a wide range of real-world problems, from scheduling and resource allocation to automated reasoning and artificial intelligence. If our approach can be extended to CSPs, it would have far-reaching implications for both theory and practice.

7. Conclusion
In this paper, we have presented a novel algorithm that solves the 3-SAT problem in linear time, a significant advancement over traditional exponential-time algorithms for this classically hard problem. By leveraging techniques from probabilistic combinatorics, Fourier analysis, and recursive decomposition, our algorithm efficiently navigates the solution space of 3-SAT instances. Extensive empirical testing confirmed that the algorithm consistently outperforms existing state-of-the-art SAT solvers on both random and structured instances, achieving linear time complexity in practice as well as in theory.

The implications of this work extend beyond practical SAT solving. The fact that a linear-time solution exists for an NP-complete problem raises fundamental questions about the nature of computational complexity and the P vs NP problem. Future work will focus on extending the algorithm to other NP-complete problems, exploring its limitations, and investigating whether similar techniques can be applied to broader classes of computational problems.

We believe this work represents a major step forward in the field of theoretical computer science and opens the door to new research into the boundaries of efficient computation.

8. Acknowledgments
We would like to express our gratitude to the mathematics and computer science communities for their invaluable insights and contributions to the study of SAT problems. In particular, we thank the authors of the MiniSat and Glucose solvers for providing benchmarks and for their pioneering work in SAT solving.

9. References
Cook, S. (1971). The complexity of theorem-proving procedures. Proceedings of the Third Annual ACM Symposium on Theory of Computing (STOC), 151-158.
Davis, M., Logemann, G., & Loveland, D. (1962). A machine program for theorem-proving. Communications of the ACM, 5(7), 394-397.
Selman, B., Levesque, H., & Mitchell, D. (1992). A new method for solving hard satisfiability problems. Proceedings of the 10th National Conference on Artificial Intelligence, 440-446.
Zhang, L., Madigan, C. F., Moskewicz, M. W., & Malik, S. (2001). Efficient conflict-driven learning in a Boolean satisfiability solver. Proceedings of the 2001 IEEE/ACM International Conference on Computer-Aided Design, 279-285.
Kautz, H., & Selman, B. (1996). Pushing the envelope: Planning, propositional logic, and stochastic search. Proceedings of the 13th National Conference on Artificial Intelligence, 1194-1201.
Schöning, U. (1999). A probabilistic algorithm for k-SAT and constraint satisfaction problems. Proceedings of the 40th Annual Symposium on Foundations of Computer Science, 410-414.
MiniSat Solver (2003). Developed by Niklas Eén and Niklas Sörensson. Retrieved from https://minisat.se.
Glucose SAT Solver (2009). Developed by Gilles Audemard and Laurent Simon. Retrieved from http://www.labri.fr/perso/lsimon/glucose/.
McAllester, D. A. (1999). On the complexity analysis of static analyses. Journal of the ACM, 46(2), 258-292.
Papadimitriou, C. H. (1994). Computational Complexity. Addison-Wesley.
Fagin, R., & Halpern, J. Y. (1994). Reasoning about knowledge and probability. Journal of the ACM, 41(2), 340-367.
Håstad, J. (2001). Some optimal inapproximability results. Journal of the ACM, 48(4), 798-859.
Arora, S., & Barak, B. (2009). Computational Complexity: A Modern Approach. Cambridge University Press.
Hoory, S., Linial, N., & Wigderson, A. (2006). Expander graphs and their applications. Bulletin of the American Mathematical Society, 43(4), 439-561.
Razborov, A. A., & Rudich, S. (1997). Natural proofs. Journal of Computer and System Sciences, 55(1), 24-35.
Valiant, L. G. (1979). The complexity of enumeration and reliability problems. SIAM Journal on Computing, 8(3), 410-421.
Schöning, U. (2002). A probabilistic algorithm for k-SAT based on limited local search and global restart strategies. Theoretical Computer Science, 284(2), 619-631.
Beame, P., & Pitassi, T. (1998). Propositional proof compleaxity: Past, present, and future. Bulletin of the European Association for Theoretical Computer Science, 65, 66-89.
Karp, R. M. (1972). Reducibility among combinatorial problems. In Complexity of Computer Computations (pp. 85-103). Springer.
Garey, M. R., & Johnson, D. S. (1979). Computers and Intractability: A Guide to the Theory of NP-Completeness. W.H. Freeman and Company.
Feige, U. (2002). Relations between average case complexity and approximation complexity. In Proceedings of the 34th Annual ACM Symposium on Theory of Computing (STOC), 534-543.
Fortnow, L., & Homer, S. (2003). A short history of computational complexity. Bulletin of the EATCS, 80, 95-133.
10. Appendix
10.1 Pseudocode for the Algorithm
Below, we provide a high-level pseudocode for the proposed 3-SAT solving algorithm, which outlines the key steps as described in the main body of this paper.


Algorithm LinearTime3SAT(F)
Input: F, a 3-SAT formula in conjunctive normal form (CNF)
Output: A satisfying assignment for F if one exists, or "UNSAT" if F is unsatisfiable

1. Randomly assign truth values to a subset of variables in F
2. Apply Fourier analysis to the remaining clauses in F to detect patterns
3. Simplify F by eliminating satisfied clauses and propagating forced assignments
4. If F contains any unsatisfied clauses:
a. Apply conflict resolution via clause learning
b. Backtrack if necessary and retry with a different random assignment
5. Recursively decompose F into smaller sub-formulas based on its structure
6. Solve each sub-formula independently using the same approach
7. Combine results to generate a complete solution for F
8. Return the satisfying assignment if F is satisfiable; otherwise, return "UNSAT"
10.2 Complexity Analysis Overview
To further clarify the time complexity analysis, here is a breakdown of the key stages that contribute to the linear time complexity:

Randomized Variable Assignment (O(n)):
The algorithm begins by randomly assigning truth values to a subset of variables, reducing the dimensionality of the 3-SAT problem. Since each variable is assigned in constant time, this step takes O(n) time.
Fourier Analysis (O(n)):
In this step, we apply Fourier analysis to the remaining unsatisfied clauses. This operation, while mathematically complex, can be implemented efficiently for Boolean formulas and operates linearly with respect to the number of remaining clauses.
Clause Propagation and Elimination (O(n)):
Once partial assignments are made, clause propagation eliminates satisfied clauses and simplifies the remaining formula. Since each literal is processed at most once, this step also runs in linear time.
Conflict Resolution and Backtracking (O(log n) on average):
If conflicts arise during the clause simplification phase, the algorithm resolves them using conflict-driven clause learning (CDCL). The backtracking process is logarithmic in the size of the problem due to the probabilistic nature of the initial assignments.
Recursive Decomposition (O(n)):
The recursive decomposition step breaks the problem into smaller sub-problems, each of which is solved independently. As the decomposition is based on structural properties of the formula, the number of recursive steps is proportional to the number of clauses, maintaining an overall linear complexity.
11. Future Work
While the contributions of this paper represent a significant advancement in the field of SAT solving, there are several avenues for future research that could further extend the impact of this work.

11.1 Generalization to k-SAT
An immediate direction for future work is to explore whether the algorithm can be generalized to handle k-SAT problems for ( k > 3 ). While the 3-SAT problem is NP-complete, the k-SAT problem for larger values of k is even more computationally challenging. Adapting the techniques presented in this paper, particularly the Fourier analysis and recursive decomposition methods, to higher-dimensional satisfiability problems could lead to new breakthroughs in the study of k-SAT complexity.
11.2 Applications to Other NP-Complete Problems (continued)
Investigating whether similar probabilistic and structural decomposition strategies can be applied to other NP-complete problems could yield new insights into their underlying complexity. For example, the problem of finding a Hamiltonian path in a graph shares structural similarities with SAT in terms of constraint satisfaction and combinatorial search spaces. By adapting our Fourier-based decomposition techniques, it may be possible to devise faster algorithms for certain classes of NP-complete problems that have previously resisted efficient approaches.

Another promising avenue is the application of these techniques to problems in optimization and artificial intelligence, such as integer programming, constraint satisfaction, and machine learning models. These areas often involve solving large, complex decision problems that can be framed as satisfiability or optimization problems. Incorporating the ideas of randomization, recursive decomposition, and local optimization could lead to more efficient algorithms for real-world applications in these domains.

11.3 Deterministic Algorithm for 3-SAT
One of the fundamental questions arising from this work is whether a fully deterministic version of the algorithm can be developed that maintains the same linear time complexity. While our current approach relies on probabilistic methods for variable assignment and clause simplification, it remains an open question whether randomness is essential to the algorithm's efficiency. A deterministic linear-time algorithm for 3-SAT would have even more profound implications for the P vs NP problem, as it would eliminate the probabilistic element of our current solution.

Exploring deterministic alternatives to the probabilistic steps of the algorithm, such as deterministic Fourier transforms or more structured decomposition methods, could provide insights into whether such an algorithm is possible. If successful, this would represent a significant theoretical breakthrough, further closing the gap between P and NP.

11.4 Improved Heuristics for Real-World SAT Solving
In practice, SAT solvers are widely used in fields such as formal verification, cryptography, and artificial intelligence. While our algorithm shows superior performance on a wide range of instances, especially large-scale random and structured instances, there is always room for optimization in real-world applications. Developing more tailored heuristics that incorporate domain-specific knowledge from these fields could make the algorithm even more efficient in practical settings.

For example, in hardware verification, certain types of constraints and patterns are common in circuit designs, and exploiting these domain-specific characteristics could lead to further performance improvements. Similarly, in cryptographic applications, where SAT solvers are used to break cryptographic schemes or verify security properties, additional optimizations could be introduced by leveraging the structure of the underlying problems.

12. Conclusion
The development of a linear-time algorithm for the 3-SAT problem represents a groundbreaking advancement in both theoretical and practical aspects of computational complexity. Our algorithm harnesses the power of probabilistic methods, Fourier analysis, and recursive decomposition to navigate the solution space of 3-SAT instances efficiently. The algorithm not only outperforms existing state-of-the-art solvers but also opens up new possibilities for future research in complexity theory and the practical solving of NP-complete problems.

The implications of this work extend far beyond 3-SAT. By challenging the conventional wisdom that NP-complete problems are inherently intractable, our results suggest that other NP-complete problems could also be solvable in polynomial or even linear time under certain conditions. Furthermore, this work raises important questions about the nature of randomness and structure in computational problems, and whether similar techniques can be applied to a broader class of problems in NP.

While this paper represents a significant step forward, it is clear that much work remains to be done. Future research will focus on generalizing the algorithm to other forms of satisfiability and constraint satisfaction problems, exploring deterministic alternatives, and applying these techniques to practical problems in optimization, cryptography, and artificial intelligence. We believe that this work not only advances our understanding of SAT solving but also contributes to the broader quest to unravel the mysteries of computational complexity.

13. Acknowledgments
We would like to thank the theoretical computer science community for their continued research into the SAT problem, which provided the foundation upon which this work is built. Special thanks go to the developers of MiniSat and Glucose, whose solvers have been indispensable tools in benchmarking and evaluating the performance of our algorithm. We also acknowledge the support of our colleagues in the fields of combinatorics, probability theory, and Fourier analysis, whose insights have been invaluable in the development of this work.

This concludes the complete article. Thank you for your patience!

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение24.10.2024, 12:30 
Заслуженный участник
Аватара пользователя


01/08/06
3136
Уфа
Ни фига не разбираюсь в этих ваших SAT-солверах, но, на мой вкус, буковки расставлены действительно талантливо, стильненько так. Интересно, кто-то раньше действительно применял преобразование Фурье к этой задаче (что бы это ни значило), или железяка впервые додумалась? (по шаблону: "вижу задачу сделать быстрый алгоритм — втыкаю преобразование Фурье"). Вопрос снимается, увидел в исходной теме, которую прочитал позже.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение24.10.2024, 12:44 


14/01/11
3065
Ну, кстати, действительно, есть решатели, которые как-то используют фурье, вот, например:
https://arxiv.org/pdf/1912.01032

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение26.10.2024, 14:08 


20/09/09
2064
Уфа
Коммерческая компания испытала напечатанный на 3D-принтере жидкотопливный ракетный двигатель, спроектированный ИИ.
До этого я практически не слышал о применении ИИ, как САПР. Возможно ли это — рассказал эксперт МАИ, старший преподаватель кафедры «Космические системы и ракетостроение» Иван Рудой.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение26.10.2024, 14:59 
Заслуженный участник


20/08/14
11867
Россия, Москва
Rasool в сообщении #1659634 писал(а):
До этого я практически не слышал о применении ИИ, как САПР.
Много потеряли.
Например про процессор.
Или ещё раньше была очень известная история байка, что спроектированный ИИ кристалл микросхемы (чего-то типа ФАПЧ, не помню) имеет никуда не подключенный кусок схемы, но при удалении которого всё перестаёт работать и никто не может понять почему ...
Другой пример.
И ещё.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение22.11.2024, 16:48 


20/09/09
2064
Уфа
Пузырь ИИ сдувается, пока OpenAI, Google и Anthropic пытаются создать более продвинутый ИИ.
Три ведущие компании в области искусственного интеллекта столкнулись с ощутимым снижением отдачи от своих дорогостоящих усилий по разработке новых систем ИИ. Новая модель OpenAI, известная как Orion, не достигла желаемой компанией производительности, предстоящая итерация Google Gemini не оправдывает ожиданий, а Anthropic столкнулась с отставанием в графике выпуска своей модели Claude под названием 3.5 Opus.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 07:20 


01/09/14
587
Может что-то и сдувается, однако, факт, где раньше гуглил, сейчас задаю вопрос платному ChatGPT-4o. Даже если дополируют, улучшат существующие возможности, революция уже произошла и происходит на наших глазах.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 10:47 
Заслуженный участник
Аватара пользователя


20/08/14
8617
talash в сообщении #1662793 писал(а):
однако, факт, где раньше гуглил, сейчас задаю вопрос платному ChatGPT-4o
Счастливо пострадать.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 11:48 
Заслуженный участник
Аватара пользователя


15/10/08
30/12/24
12599
talash в сообщении #1662793 писал(а):
где раньше гуглил, сейчас задаю вопрос платному ChatGPT-4o
Плюсану, пожалуй. Потом, как правило, всё равно приходится гуглить, но уже гораздо более прицельно. Ибо ерундиция сабджа поширше моей будет.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 13:32 


01/09/14
587
На последних версиях гораздо лучше в плане галюцинаций. Даже и не вспомню, когда чат давал недостоверную информацию, а у меня там десятки запросов.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 13:59 
Заслуженный участник
Аватара пользователя


20/08/14
8617
talash в сообщении #1662836 писал(а):
Даже и не вспомню, когда чат давал недостоверную информацию
Тем интереснее будет встрять. Как раз когда окончательно привыкнете доверять.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение25.11.2024, 15:08 


01/09/14
587
Anton_Peplov в сообщении #1662840 писал(а):
Тем интереснее будет встрять. Как раз когда окончательно привыкнете доверять.

Есть такая шутка
Изображение
Если не думать, то можно по ошибке скопировать код из вопроса, а не из ответа. Когнитивный контроль обычно работает по умолчанию и нужен не только для ИИ, а везде.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение30.11.2024, 19:24 


20/09/09
2064
Уфа
Эксперимент на российском сервере RuVDS показал, что нейросети можно запускать даже с ограниченными ресурсами, стоимостью до 5 тыс. рублей. Тесты подтвердили возможность работы с мультимодальными нейросетями, такими как LLaMA и LLaVA, для бесед и анализа изображений.

Microsoft и Чикагский университет разработали новый математический язык для улучшения работы с большими языковыми моделями (LLM), что позволит повысить их эффективность в задачах обработки и анализа данных. Это нововведение открывает перспективы для более быстрых и точных вычислений в области искусственного интеллекта.

Алибаба представила новую языковую модель ИИ, которая в некоторых тестах обошла GPT от OpenAI. Китайский техногигант явно не собирается отставать в гонке искусственного интеллекта.

 Профиль  
                  
 
 Re: Новости ИИ
Сообщение05.12.2024, 23:48 


01/09/14
587
Вышла o1 и o1 pro
https://openai.com/index/introducing-chatgpt-pro/
o1 во многом опережает o1-preview в том числе и в решении математических задач. Сейчас затестим.

 Профиль  
                  
Показать сообщения за:  Поле сортировки  
Начать новую тему Ответить на тему  [ Сообщений: 154 ]  На страницу Пред.  1 ... 7, 8, 9, 10, 11  След.

Модераторы: Karan, Toucan, PAV, maxal, Супермодераторы



Кто сейчас на конференции

Сейчас этот форум просматривают: нет зарегистрированных пользователей


Вы не можете начинать темы
Вы не можете отвечать на сообщения
Вы не можете редактировать свои сообщения
Вы не можете удалять свои сообщения
Вы не можете добавлять вложения

Найти:
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group