arXiv:2505.10361v2 Announce Type: replace-cross 
Abstract: Agents are minimally entities that are influenced by their past observations and act to influence future observations. This latter capacity is captured by empowerment, which has served as a vital framing concept across artificial intelligence and cognitive science. This former capacity, however, is equally foundational: In what ways, and to what extent, can an agent be influenced by what it observes? In this paper, we ground this concept in a universal agent-centric measure that we refer to as plasticity, and reveal a fundamental connection to empowerment. Following a set of desiderata on a suitable definition, we define plasticity using a new information-theoretic quantity we call the generalized directed information. We show that this new quantity strictly generalizes the directed information introduced by Massey (1990) while preserving all of its desirable properties. Under this definition, we find that plasticity is well thought of as the mirror of empowerment: The two concepts are defined using the same measure, with only the direction of influence reversed. Our main result establishes a tension between the plasticity and empowerment of an agent, suggesting that agent design needs to be mindful of both characteristics. We explore the implications of these findings, and suggest that plasticity, empowerment, and their relationship are essential to understanding agency

تستكشف الورقة المعنونة 'Plasticity as the Mirror of Empowerment' الدور المزدوج للوكيل في الذكاء الاصطناعي وعلم النفس المعرفي، مع التركيز على كيفية تأثرهم بالملاحظات السابقة وكيف يمكنهم التأثير على الملاحظات المستقبلية. تعتبر هذه الأبحاث مهمة لأنها تعمق فهمنا للتمكين في الذكاء الاصطناعي، مما يبرز الجوانب الأساسية للملاحظة والتأثير، وهي ضرورية لتطوير أنظمة ذكاء اصطناعي أكثر تطوراً واستجابة.

El artículo titulado 'Plasticity as the Mirror of Empowerment' explora el doble papel de los agentes en la inteligencia artificial y la ciencia cognitiva, centrándose en cómo son influenciados por observaciones pasadas y cómo pueden influir en las futuras. Esta investigación es significativa ya que profundiza nuestra comprensión del empoderamiento en la IA, destacando los aspectos fundamentales de la observación y la influencia, que son cruciales para desarrollar sistemas de IA más sofisticados y receptivos.

L'article intitulé 'Plasticity as the Mirror of Empowerment' examine le double rôle des agents dans l'intelligence artificielle et la science cognitive, en se concentrant sur la manière dont ils sont influencés par des observations passées et comment ils peuvent influencer celles à venir. Cette recherche est importante car elle approfondit notre compréhension de l'autonomisation dans l'IA, mettant en lumière les aspects fondamentaux de l'observation et de l'influence, qui sont cruciaux pour développer des systèmes d'IA plus sophistiqués et réactifs.

The paper titled 'Plasticity as the Mirror of Empowerment' explores the dual role of agents in artificial intelligence and cognitive science, focusing on how they are influenced by past observations and how they can influence future ones. This research is significant as it deepens our understanding of empowerment in AI, highlighting the foundational aspects of observation and influence, which are crucial for developing more sophisticated and responsive AI systems.

Plasticity as the Mirror of Empowerment

arXiv:2601.08169v1 Announce Type: cross 
Abstract: Representing relations between concepts is a core prerequisite for intelligent systems to make sense of the world. Recent work using causal mediation analysis has shown that a small set of attention heads encodes task representation in in-context learning, captured in a compact representation known as the function vector. We show that fine-tuning function vectors with only a small set of examples (about 20 word pairs) yields better performance on relation-based word-completion tasks than using the original vectors derived from causal mediation analysis. These improvements hold for both small and large language models. Moreover, the fine-tuned function vectors yield improved decoding performance for relation words and show stronger alignment with human similarity judgments of semantic relations. Next, we introduce the composite function vector - a weighted combination of fine-tuned function vectors - to extract relational knowledge and support analogical reasoning. At inference time, inserting this composite vector into LLM activations markedly enhances performance on challenging analogy problems drawn from cognitive science and SAT benchmarks. Our results highlight the potential of activation patching as a controllable mechanism for encoding and manipulating relational knowledge, advancing both the interpretability and reasoning capabilities of large language models.

أظهرت الأبحاث الحديثة أن ضبط المتجهات الوظيفية، المستمدة من تحليل الوساطة السببية، يمكن أن يحسن بشكل كبير الأداء في مهام إكمال الكلمات المعتمدة على العلاقات. من خلال استخدام مجموعة صغيرة من الأمثلة، حوالي 20 زوجًا من الكلمات، تكشف الدراسة أن هذه المتجهات المضبوطة تتفوق على المتجهات الأصلية في كل من نماذج اللغة الصغيرة والكبيرة.

Investigaciones recientes han demostrado que el ajuste fino de vectores de función, derivados del análisis de mediación causal, puede mejorar significativamente el rendimiento en tareas de completado de palabras basadas en relaciones. Al utilizar un pequeño conjunto de ejemplos, aproximadamente 20 pares de palabras, el estudio revela que estos vectores ajustados superan a los vectores originales en modelos de lenguaje tanto pequeños como grandes.

Des recherches récentes ont démontré que le réglage fin des vecteurs de fonction, dérivés de l'analyse de médiation causale, peut améliorer considérablement les performances dans les tâches de complétion de mots basées sur les relations. En utilisant un petit ensemble d'exemples, environ 20 paires de mots, l'étude révèle que ces vecteurs ajustés surpassent les vecteurs originaux dans les modèles de langage petits et grands.

Recent research has demonstrated that fine-tuning function vectors, derived from causal mediation analysis, can significantly enhance performance in relation-based word-completion tasks. By utilizing a small set of examples, approximately 20 word pairs, the study reveals that these fine-tuned vectors outperform original vectors in both small and large language models.

Relational Knowledge Distillation Using Fine-tuned Function Vectors

arXiv:2411.00365v2 Announce Type: replace 
Abstract: In the paradigm of decentralized learning, a group of agents collaborate to learn a global model using a distributed dataset without a central server; nevertheless, it is severely challenged by the heterogeneity of the data distribution across the agents. For example, the data may be distributed non-independently and identically, and even be noised or poisoned. To address these data challenges, we propose ROSS, a novel robust decentralized stochastic learning algorithm based on Shapley values, in this paper. Specifically, in each round, each agent aggregates the cross-gradient information from its neighbors, i.e., the derivatives of its local model with respect to the datasets of its neighbors, to update its local model in a momentum like manner, while we innovate in weighting the derivatives according to their contributions measured by Shapley values. We perform solid theoretical analysis to reveal the linear convergence speedup of our ROSS algorithm. We also verify the efficacy of our algorithm through extensive experiments on public datasets. Our results demonstrate that, in face of the above variety of data challenges, our ROSS algorithm has significant advantages over existing state-of-the-art proposals in terms of both convergence and prediction accuracy.

تم اقتراح خوارزمية جديدة للتعلم اللامركزي تُدعى ROSS، والتي تستخدم قيم شابلي لتعزيز قوة التعلم العشوائي بين الوكلاء. تتناول هذه الطريقة التحديات التي تطرحها توزيعات البيانات غير المتجانسة، مما يسمح للوكلاء بتعلم نموذج عالمي بشكل تعاوني دون الحاجة إلى خادم مركزي. يقوم كل وكيل بتحديث نموذجه من خلال تجميع معلومات التدرج المتقاطع من الوكلاء المجاورين، مع وزنها وفقًا لمساهماتهم.

Se ha propuesto un nuevo algoritmo de aprendizaje descentralizado llamado ROSS, que utiliza valores de Shapley para mejorar la robustez del aprendizaje estocástico entre agentes. Este enfoque aborda los desafíos planteados por las distribuciones de datos heterogéneas, permitiendo que los agentes aprendan de manera colaborativa un modelo global sin un servidor central. Cada agente actualiza su modelo agregando información de gradiente cruzado de los agentes vecinos, ponderada por sus contribuciones.

Un nouvel algorithme d'apprentissage décentralisé nommé ROSS a été proposé, utilisant les valeurs de Shapley pour améliorer la robustesse de l'apprentissage stochastique parmi les agents. Cette approche répond aux défis posés par les distributions de données hétérogènes, permettant aux agents d'apprendre de manière collaborative un modèle global sans serveur central. Chaque agent met à jour son modèle en agrégeant des informations de gradient croisé provenant des agents voisins, pondérées par leurs contributions.

A new decentralized learning algorithm named ROSS has been proposed, which utilizes Shapley values to enhance the robustness of stochastic learning among agents. This approach addresses challenges posed by heterogeneous data distributions, allowing agents to collaboratively learn a global model without a central server. Each agent updates its model by aggregating cross-gradient information from neighboring agents, weighted by their contributions.

ROSS: RObust decentralized Stochastic learning based on Shapley values

One More Thing in AI – Your Shortcut to AI Mastery

Plasticity as the Mirror of Empowerment

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

Chattermate

Teilord

AIvilization

Emergent.sh

Ready to build your own newsroom?