Search

We study the ground state energy of trapped two-dimensional Bose gases with mean-field type interactions. We prove the stability of second kind of the many-body system and the convergence of the ground state energy per particle to that of a non-linear Schr\"odinger (NLS) energy functional. Notably, we can take the scaling of the interaction to be arbitrarily close to the Gross-Pitaevskii scaling, which is a drastic improvement on the best-known result for systems with attractive interactions. As a consequence of the stability of second kind we also obtain Bose-Einstein condensation for the many-body ground states and dynamics for a much improved range of the diluteness parameter., Comment: 16 pages

To advance autonomous dexterous manipulation, we propose a hybrid control method that combines the relative advantages of a fine-tuned Vision-Language-Action (VLA) model and diffusion models. The VLA model provides language commanded high-level planning, which is highly generalizable, while the diffusion model handles low-level interactions which offers the precision and robustness required for specific objects and environments. By incorporating a switching signal into the training-data, we enable event based transitions between these two models for a pick-and-place task where the target object and placement location is commanded through language. This approach is deployed on our anthropomorphic ADAPT Hand 2, a 13DoF robotic hand, which incorporates compliance through series elastic actuation allowing for resilience for any interactions: showing the first use of a multi-fingered hand controlled with a VLA model. We demonstrate this model switching approach results in a over 80\% success rate compared to under 40\% when only using a VLA model, enabled by accurate near-object arm motion by the VLA model and a multi-modal grasping motion with error recovery abilities from the diffusion model.

Reward models (RM) play a critical role in aligning generations of large language models (LLM) to human expectations. However, prevailing RMs fail to capture the stochasticity within human preferences and cannot effectively evaluate the reliability of reward predictions. To address these issues, we propose Uncertain-aware RM (URM) and Uncertain-aware RM Ensemble (URME) to incorporate and manage uncertainty in reward modeling. URM can model the distribution of disentangled attributes within human preferences, while URME quantifies uncertainty through discrepancies in the ensemble, thereby identifying potential lack of knowledge during reward evaluation. Experiment results indicate that the proposed URM achieves state-of-the-art performance compared to models with the same size, demonstrating the effectiveness of modeling uncertainty within human preferences. Furthermore, empirical results show that through uncertainty quantification, URM and URME can identify unreliable predictions to improve the quality of reward evaluations.

Large Language Models (LLMs) have revolutionized artificial intelligence and machine learning through their advanced text processing and generating capabilities. However, their widespread deployment has raised significant safety and reliability concerns. Established vulnerabilities in deep neural networks, coupled with emerging threat models, may compromise security evaluations and create a false sense of security. Given the extensive research in the field of LLM security, we believe that summarizing the current state of affairs will help the research community better understand the present landscape and inform future developments. This paper reviews current research on LLM vulnerabilities and threats, and evaluates the effectiveness of contemporary defense mechanisms. We analyze recent studies on attack vectors and model weaknesses, providing insights into attack mechanisms and the evolving threat landscape. We also examine current defense strategies, highlighting their strengths and limitations. By contrasting advancements in attack and defense methodologies, we identify research gaps and propose future directions to enhance LLM security. Our goal is to advance the understanding of LLM safety challenges and guide the development of more robust security measures.

We consider a dilute Bose gas in the thermodynamic limit and prove a lower bound on the free energy for low temperatures which is in agreement with the conjecture of Lee-Huang-Yang on the excitation spectrum of the system. Combining techniques of \cite{FS2} and \cite{HHNST}, we give a simpler and shorter proof resolving the case of strong interactions, including the hard-core potential., Comment: 33 pages

Simulating quantum dynamics is one of the most promising applications of quantum computers. While the upper bound of the simulation cost has been extensively studied through various quantum algorithms, much less work has focused on establishing the lower bound, particularly for the simulation of open quantum system dynamics. In this work, we present a general framework to calculate the lower bound for simulating a broad class of quantum Markov semigroups. Given a fixed accessible unitary set, we introduce the concept of convexified circuit depth to quantify the quantum simulation cost and analyze the necessary circuit depth to construct a quantum simulation scheme that achieves a specific order. Our framework can be applied to both unital and non-unital quantum dynamics, and the tightness of our lower bound technique is illustrated by showing that the upper and lower bounds coincide in several examples.

We consider a gas of bosons interacting through a three-body hard-core potential in the thermodynamic limit. We derive an upper bound on the ground state energy of the system at the leading order using a Jastrow factor. Our result matches the lower bound proven by Nam-Ricaud-Triay and therefore resolves the leading order. Moreover, a straightforward adaptation of our proof can be used for systems interacting via combined two-body and three-body interactions to generalise Theorem 1.2. from arXiv:2402.05646 to hard-core potentials.

We show the failure of a matricial version of Grothendieck's theorem for operator spaces, thereby resolving a long-standing open question in the field. Moreover, by showing that such a counterexample can occur in the simplest context of commutative $C^*$-algebras, we address some other open questions in operator algebras. Our constructions, completely explicit and fairly simple, are inspired by some techniques in quantum information theory., Comment: v2 Minor modifications. Same results

Bak, Tang, and Wiesenfeld developed their theory of self-organized criticality in the late 1980s to explain why many real-life processes exhibit signs of critical behavior despite the absence of a tuning parameter. A decade later, Dickman, Mu\~noz, Vespignani, and Zapperi explained self-organized criticality as an external force pushing a hidden parameter toward the critical value of a traditional absorbing-state phase transition. As evidence, they observed empirically that for various sandpile models, the particle density in a finite box under driven-dissipative dynamics converges to the critical density of an infinite-volume version of the model. We give the first proof of this well-known density conjecture in any setting by establishing it for activated random walk in one dimension. We prove that two other natural versions of the model have the same critical value, further establishing activated random walk as a universal model of self-organized criticality., Comment: 77 pages; split Section 3 into two sections, improved the presentation in various ways, and added new expository figures

Decision making demands intricate interplay between perception, memory, and reasoning to discern optimal policies. Conventional approaches to decision making face challenges related to low sample efficiency and poor generalization. In contrast, foundation models in language and vision have showcased rapid adaptation to diverse new tasks. Therefore, we advocate for the construction of foundation agents as a transformative shift in the learning paradigm of agents. This proposal is underpinned by the formulation of foundation agents with their fundamental characteristics and challenges motivated by the success of large language models (LLMs). Moreover, we specify the roadmap of foundation agents from large interactive data collection or generation, to self-supervised pretraining and adaptation, and knowledge and value alignment with LLMs. Lastly, we pinpoint critical research questions derived from the formulation and delineate trends for foundation agents supported by real-world use cases, addressing both technical and theoretical aspects to propel the field towards a more comprehensive and impactful future., Comment: 17 pages, camera-ready version of ICML 2024

Human preference alignment is critical in building powerful and reliable large language models (LLMs). However, current methods either ignore the multi-dimensionality of human preferences (e.g. helpfulness and harmlessness) or struggle with the complexity of managing multiple reward models. To address these issues, we propose Sequential Preference Optimization (SPO), a method that sequentially fine-tunes LLMs to align with multiple dimensions of human preferences. SPO avoids explicit reward modeling, directly optimizing the models to align with nuanced human preferences. We theoretically derive closed-form optimal SPO policy and loss function. Gradient analysis is conducted to show how SPO manages to fine-tune the LLMs while maintaining alignment on previously optimized dimensions. Empirical results on LLMs of different size and multiple evaluation datasets demonstrate that SPO successfully aligns LLMs across multiple dimensions of human preferences and significantly outperforms the baselines.

The impressive capabilities of humans to robustly perform manipulation relies on compliant interactions, enabled through the structure and materials spatially distributed in our hands. We propose by mimicking this distributed compliance in an anthropomorphic robotic hand, the open-loop manipulation robustness increases and observe the emergence of human-like behaviours. To achieve this, we introduce the ADAPT Hand equipped with tunable compliance throughout the skin, fingers, and the wrist. Through extensive automated pick-and-place tests, we show the grasping robustness closely mirrors an estimated geometric theoretical limit, while `stress-testing' the robot hand to perform 800+ grasps. Finally, 24 items with largely varying geometries are grasped in a constrained environment with a success rate of 93%. We demonstrate the hand-object self-organization behavior underlines this extreme robustness, where the hand automatically exhibits different grasp types depending on object geometries. Furthermore, the robot grasp type mimics a natural human grasp with a direct similarity of 68%.

Generating unbounded 3D scenes is crucial for large-scale scene understanding and simulation. Urban scenes, unlike natural landscapes, consist of various complex man-made objects and structures such as roads, traffic signs, vehicles, and buildings. To create a realistic and detailed urban scene, it is crucial to accurately represent the geometry and semantics of the underlying objects, going beyond their visual appearance. In this work, we propose UrbanDiffusion, a 3D diffusion model that is conditioned on a Bird's-Eye View (BEV) map and generates an urban scene with geometry and semantics in the form of semantic occupancy map. Our model introduces a novel paradigm that learns the data distribution of scene-level structures within a latent space and further enables the expansion of the synthesized scene into an arbitrary scale. After training on real-world driving datasets, our model can generate a wide range of diverse urban scenes given the BEV maps from the held-out set and also generalize to the synthesized maps from a driving simulator. We further demonstrate its application to scene image synthesis with a pretrained image generator as a prior., Comment: The project website is https://metadriverse.github.io/urbandiff/

In the development of advanced Texas Hold'em AI systems, abstraction technology has garnered widespread attention due to its significant effect in simplifying game complexity. This study adopts a more specific model, the games of ordered signal, to describe Texas Hold'em-style games and optimizes this model to streamline its mathematical representation and broaden its applicability. By transitioning from a broad imperfect information game model to a game with ordered signals model, we have separated the previously intertwined infoset abstraction and action abstraction into independent signal abstraction and action abstraction. Importantly, this signal abstraction provides a mathematical framework for the hand abstraction task, which is emphatically discussed in this paper. Additionally, a novel common refinement principle is introduced, revealing the limit performance of hand abstraction algorithms. We introduce potential outcome isomorphism (POI) and pinpoint that it suffers from the issue of excessive abstraction. Futher, We demonstrate that POI serves as a common refinement for leading outcome-based hand abstraction algorithms, such as E[HS] and PA\&PAEMD. Consequently, excessive abstraction also inherently affects these algorithms, leading to suboptimal performance. Our investigation reveals the omission of historical data as a primary contributor to excessive abstraction. To remedy this, we propose the K-Recall Outcome Isomorphism (KROI) to incorporate the missing information. Compared with POI, KROI more accurately mirrors lossless isomorphism (LI), the ground truth, offering enhanced signal abstraction resolution. Experimental results in the Numeral211 Hold'em indicate that strategies developed through KROI approximate the exploitability of those developed through LI more closely than those trained through POI., Comment: 36 pages, 7 figures

Autonomous driving simulation system plays a crucial role in enhancing self-driving data and simulating complex and rare traffic scenarios, ensuring navigation safety. However, traditional simulation systems, which often heavily rely on manual modeling and 2D image editing, struggled with scaling to extensive scenes and generating realistic simulation data. In this study, we present S-NeRF++, an innovative autonomous driving simulation system based on neural reconstruction. Trained on widely-used self-driving datasets such as nuScenes and Waymo, S-NeRF++ can generate a large number of realistic street scenes and foreground objects with high rendering quality as well as offering considerable flexibility in manipulation and simulation. Specifically, S-NeRF++ is an enhanced neural radiance field for synthesizing large-scale scenes and moving vehicles, with improved scene parameterization and camera pose learning. The system effectively utilizes noisy and sparse LiDAR data to refine training and address depth outliers, ensuring high-quality reconstruction and novel-view rendering. It also provides a diverse foreground asset bank by reconstructing and generating different foreground vehicles to support comprehensive scenario creation.Moreover, we have developed an advanced foreground-background fusion pipeline that skillfully integrates illumination and shadow effects, further enhancing the realism of our simulations. With the high-quality simulated data provided by our S-NeRF++, we found the perception methods enjoy performance boosts on several autonomous driving downstream tasks, further demonstrating our proposed simulator's effectiveness.

Safe reinforcement learning (RL) agents accomplish given tasks while adhering to specific constraints. Employing constraints expressed via easily-understandable human language offers considerable potential for real-world applications due to its accessibility and non-reliance on domain expertise. Previous safe RL methods with natural language constraints typically adopt a recurrent neural network, which leads to limited capabilities when dealing with various forms of human language input. Furthermore, these methods often require a ground-truth cost function, necessitating domain expertise for the conversion of language constraints into a well-defined cost function that determines constraint violation. To address these issues, we proposes to use pre-trained language models (LM) to facilitate RL agents' comprehension of natural language constraints and allow them to infer costs for safe policy learning. Through the use of pre-trained LMs and the elimination of the need for a ground-truth cost, our method enhances safe policy learning under a diverse set of human-derived free-form natural language constraints. Experiments on grid-world navigation and robot control show that the proposed method can achieve strong performance while adhering to given constraints. The usage of pre-trained LMs allows our method to comprehend complicated constraints and learn safe policies without the need for ground-truth cost at any stage of training or evaluation. Extensive ablation studies are conducted to demonstrate the efficacy of each part of our method.

Decision-making is a dynamic process requiring perception, memory, and reasoning to make choices and find optimal policies. Traditional approaches to decision-making suffer from sample efficiency and generalization, while large-scale self-supervised pretraining has enabled fast adaptation with fine-tuning or few-shot learning in language and vision. We thus argue to integrate knowledge acquired from generic large-scale self-supervised pretraining into downstream decision-making problems. We propose Pretrain-Then-Adapt pipeline and survey recent work on data collection, pretraining objectives and adaptation strategies for decision-making pretraining and downstream inference. Finally, we identify critical challenges and future directions for developing decision foundation model with the help of generic and flexible self-supervised pretraining.

Designing better deep networks and better reinforcement learning (RL) algorithms are both important for deep RL. This work studies the former. Specifically, the Perception and Decision-making Interleaving Transformer (PDiT) network is proposed, which cascades two Transformers in a very natural way: the perceiving one focuses on \emph{the environmental perception} by processing the observation at the patch level, whereas the deciding one pays attention to \emph{the decision-making} by conditioning on the history of the desired returns, the perceiver's outputs, and the actions. Such a network design is generally applicable to a lot of deep RL settings, e.g., both the online and offline RL algorithms under environments with either image observations, proprioception observations, or hybrid image-language observations. Extensive experiments show that PDiT can not only achieve superior performance than strong baselines in different settings but also extract explainable feature representations. Our code is available at \url{https://github.com/maohangyu/PDiT}., Comment: Proc. of the 23rd International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2024, full paper with oral presentation). Cover our preliminary study: arXiv:2212.14538

Multi-Agent Policy Gradient (MAPG) has made significant progress in recent years. However, centralized critics in state-of-the-art MAPG methods still face the centralized-decentralized mismatch (CDM) issue, which means sub-optimal actions by some agents will affect other agent's policy learning. While using individual critics for policy updates can avoid this issue, they severely limit cooperation among agents. To address this issue, we propose an agent topology framework, which decides whether other agents should be considered in policy gradient and achieves compromise between facilitating cooperation and alleviating the CDM issue. The agent topology allows agents to use coalition utility as learning objective instead of global utility by centralized critics or local utility by individual critics. To constitute the agent topology, various models are studied. We propose Topology-based multi-Agent Policy gradiEnt (TAPE) for both stochastic and deterministic MAPG methods. We prove the policy improvement theorem for stochastic TAPE and give a theoretical explanation for the improved cooperation among agents. Experiment results on several benchmarks show the agent topology is able to facilitate agent cooperation and alleviate CDM issue respectively to improve performance of TAPE. Finally, multiple ablation studies and a heuristic graph search algorithm are devised to show the efficacy of the agent topology.

Backdoor attacks in reinforcement learning (RL) have previously employed intense attack strategies to ensure attack success. However, these methods suffer from high attack costs and increased detectability. In this work, we propose a novel approach, BadRL, which focuses on conducting highly sparse backdoor poisoning efforts during training and testing while maintaining successful attacks. Our algorithm, BadRL, strategically chooses state observations with high attack values to inject triggers during training and testing, thereby reducing the chances of detection. In contrast to the previous methods that utilize sample-agnostic trigger patterns, BadRL dynamically generates distinct trigger patterns based on targeted state observations, thereby enhancing its effectiveness. Theoretical analysis shows that the targeted backdoor attack is always viable and remains stealthy under specific assumptions. Empirical results on various classic RL tasks illustrate that BadRL can substantially degrade the performance of a victim agent with minimal poisoning efforts 0.003% of total training steps) during training and infrequent attacks during testing., Comment: Extended version of the submission accepted by AAAI 2024. It is revised by integrating review comments

Recurrence is a fundamental characteristic of complicated deterministic dynamical systems. Understanding its structure is challenging, especially if the system of interest has many degrees of freedom so that visualizations of trajectories are of limited use. To analyze recurrent phenomena, we propose a computational method to identify oscillations and the transitions between them. To this end, we introduce the concept of cycling signature, which is a topological descriptor of a trajectory segment. The advantage of a topological approach, in particular in an application setting, is that it is robust to noise. Cycling signatures are computable from data and provide a comprehensive global description of oscillations through their statistics over many trajectory segments. We demonstrate this through three examples. In particular, we identify and analyze six oscillations in a four-dimensional system with a hyperchaotic attractor., Comment: main paper: 11 pages, 7 figures; appendix: 9 pages, 8 figures

Continual learning addresses the problem of continuously acquiring and transferring knowledge without catastrophic forgetting of old concepts. While humans achieve continual learning via diverse neurocognitive mechanisms, there is a mismatch between cognitive properties and evaluation methods of continual learning models. First, the measurement of continual learning models mostly relies on evaluation metrics at a micro-level, which cannot characterize cognitive capacities of the model. Second, the measurement is method-specific, emphasizing model strengths in one aspect while obscuring potential weaknesses in other respects. To address these issues, we propose to integrate model cognitive capacities and evaluation metrics into a unified evaluation paradigm. We first characterize model capacities via desiderata derived from cognitive properties supporting human continual learning. The desiderata concern (1) adaptability in varying lengths of task sequence; (2) sensitivity to dynamic task variations; and (3) efficiency in memory usage and training time consumption. Then we design evaluation protocols for each desideratum to assess cognitive capacities of recent continual learning models. Experimental results show that no method we consider has satisfied all the desiderata and is still far away from realizing truly continual learning. Although some methods exhibit some degree of adaptability and efficiency, no method is able to identify task relationships when encountering dynamic task variations, or achieve a trade-off in learning similarities and differences between tasks. Inspired by these results, we discuss possible factors that influence model performance in these desiderata and provide guidance for the improvement of continual learning models., Comment: 12 pages, 11 figures

Building agents with adaptive behavior in cooperative tasks stands as a paramount goal in the realm of multi-agent systems. Current approaches to developing cooperative agents rely primarily on learning-based methods, whose policy generalization depends heavily on the diversity of teammates they interact with during the training phase. Such reliance, however, constrains the agents' capacity for strategic adaptation when cooperating with unfamiliar teammates, which becomes a significant challenge in zero-shot coordination scenarios. To address this challenge, we propose ProAgent, a novel framework that harnesses large language models (LLMs) to create proactive agents capable of dynamically adapting their behavior to enhance cooperation with teammates. ProAgent can analyze the present state, and infer the intentions of teammates from observations. It then updates its beliefs in alignment with the teammates' subsequent actual behaviors. Moreover, ProAgent exhibits a high degree of modularity and interpretability, making it easily integrated into various of coordination scenarios. Experimental evaluations conducted within the Overcooked-AI environment unveil the remarkable performance superiority of ProAgent, outperforming five methods based on self-play and population-based training when cooperating with AI agents. Furthermore, in partnered with human proxy models, its performance exhibits an average improvement exceeding 10% compared to the current state-of-the-art method. For more information about our project, please visit~\url{https://pku-proagent.github.io}., Comment: v3 is the AAAI'24 camera ready version, which polished abstract and introduction based on the reviewers' comments, and enriched related works. 7 pages of main content, 2 pages of references, 2 figures and 1 table

We generalise the dynamic Laplacian introduced in (Froyland, 2015) to a dynamic $p$-Laplacian, in analogy to the generalisation of the standard $2$-Laplacian to the standard $p$-Laplacian for $p>1$. Spectral properties of the dynamic Laplacian are connected to the geometric problem of finding "coherent" sets with persistently small boundaries under dynamical evolution, and we show that the dynamic $p$-Laplacian shares similar geometric connections. In particular, we prove that the first eigenvalue of the dynamic $p$-Laplacian with Dirichlet boundary conditions exists and converges to a dynamic version of the Cheeger constant introduced in (Froyland, 2015) as $p\rightarrow 1$. We develop a numerical scheme to estimate the leading eigenfunctions of the (nonlinear) dynamic $p$-Laplacian, and through a series of examples we investigate the behaviour of the level sets of these eigenfunctions. These level sets define the boundaries of sets in the domain of the dynamics that remain coherent under the dynamical evolution.

Deep reinforcement learning algorithms are usually impeded by sampling inefficiency, heavily depending on multiple interactions with the environment to acquire accurate decision-making capabilities. In contrast, humans rely on their hippocampus to retrieve relevant information from past experiences of relevant tasks, which guides their decision-making when learning a new task, rather than exclusively depending on environmental interactions. Nevertheless, designing a hippocampus-like module for an agent to incorporate past experiences into established reinforcement learning algorithms presents two challenges. The first challenge involves selecting the most relevant past experiences for the current task, and the second challenge is integrating such experiences into the decision network. To address these challenges, we propose a novel method that utilizes a retrieval network based on task-conditioned hypernetwork, which adapts the retrieval network's parameters depending on the task. At the same time, a dynamic modification mechanism enhances the collaborative efforts between the retrieval and decision networks. We evaluate the proposed method across various tasks within a multitask scenario in the Minigrid environment. The experimental results demonstrate that our proposed method significantly outperforms strong baselines.

This thesis provides an in-depth exploration of the Decentralized Co-governance Crowdfunding (DCC) Ecosystem, a novel solution addressing prevailing challenges in conventional crowdfunding methods faced by MSMEs and innovative projects. Among the problems it seeks to mitigate are high transaction costs, lack of transparency, fraud, and inefficient resource allocation. Leveraging a comprehensive review of the existing literature on crowdfunding economic activities and blockchain's impact on organizational governance, we propose a transformative socio-economic model based on digital tokens and decentralized co-governance. This ecosystem is marked by a tripartite community structure - the Labor, Capital, and Governance communities - each contributing uniquely to the ecosystem's operation. Our research unfolds the evolution of the DCC ecosystem through distinct phases, offering a novel understanding of socioeconomic dynamics in a decentralized digital world. It also delves into the intricate governance mechanism of the ecosystem, ensuring integrity, fairness, and a balanced distribution of value and wealth.