ICML 2024 Papers

The combination of self-play and planning has achieved great successes in sequential games, for instance in Chess and Go. However, adapting algorithms such as AlphaZero to simultaneous games poses a new challenge. In these games, missing information about concurrent actions of other agents is a limiting factor as they may select different Nash equilibria or do not play optimally at all. Thus, it is vital to model the behavior of the other agents when interacting with them in simultaneous games. To this end, we propose Albatross: AlphaZero for Learning Bounded-rational Agents and Temperature-based Response Optimization using Simulated Self-play. Albatross learns to play the novel equilibrium concept of a Smooth Best Response Logit Equilibrium (SBRLE), which enables cooperation and competition with agents of any playing strength. We perform an extensive evaluation of Albatross on a set of cooperative and competitive simultaneous perfect-information games. In contrast to AlphaZero, Albatross is able to exploit weak agents in the competitive game of Battlesnake. Additionally, it yields an improvement of 37.6% compared to previous state of the art in the cooperative Overcooked benchmark.

Implicit Neural Representations have gained prominence as a powerful framework for capturing complex data modalities, encompassing a wide range from 3D shapes to images and audio. Within the realm of 3D shape representation, Neural Signed Distance Functions (SDF) have demonstrated remarkable potential in faithfully encoding intricate shape geometry. However, learning SDFs from sparse 3D point clouds in the absence of ground truth supervision remains a very challenging task. While recent methods rely on smoothness priors to regularize the learning, our method introduces a regularization term that leverages adversarial samples around the shape to improve the learned SDFs. Through extensive experiments and evaluations, we illustrate the efficacy of our proposed method, highlighting its capacity to improve SDF learning with respect to baselines and the state-of-the-art using synthetic and real data.

Recent successes suggest that parameter-efficient fine-tuning of foundation models is becoming the state-of-the-art method for transfer learning in vision, gradually replacing the rich literature of alternatives such as meta-learning. In trying to harness the best of both worlds, meta-tuning introduces a subsequent optimization stage of foundation models but has so far only shown limited success and crucially tends to underperform on out-of-distribution (OOD) tasks. In this paper, we introduce Sparse MetA-Tuning (SMAT), a method inspired by sparse mixture-of-experts approaches and trained to isolate subsets of pre-trained parameters automatically for meta-tuning on each task. SMAT successfully overcomes OOD sensitivity and delivers on the promise of enhancing the transfer abilities of vision foundation models beyond parameter-efficient finetuning. We establish new state-of-the-art results on a challenging combination of Meta-Dataset augmented with additional OOD tasks in both zero-shot and gradient-based adaptation settings. In addition, we provide a thorough analysis of the superiority of learned over hand-designed sparsity patterns for sparse expert methods and the pivotal importance of the sparsity level in balancing between in-distribution and out-of-distribution generalization. Our code and models are publicly available.

The goal of strategic classification is to learn decision rules which are robust to strategic input manipulation. Earlier works assume that these responses are known; while some recent works handle unknown responses, they exclusively study online settings with repeated model deployments. But there are many domains – particularly in public policy, a common motivating use case – where multiple deployments are infeasible, or where even one bad round is unacceptable. To address this gap, we initiate the formal study of *one-shot* strategic classification under unknown responses, which requires committing to a single classifier once. Focusing on uncertainty in the users' cost function, we begin by proving that for a broad class of costs, even a small mis-estimation of the true cost can entail trivial accuracy in the worst case. In light of this, we frame the task as a minimax problem, aiming to minimize worst-case risk over an uncertainty set of costs. We design efficient algorithms for both the full-batch and stochastic settings, which we prove converge (offline) to the minimax solution at the rate of $\tilde{\mathcal{O}}(T^{-\frac{1}{2}})$. Our analysis reveals important structure stemming from strategic responses, particularly the value of *dual norm regularization* with respect to the cost function.

Subset or core-set selection offers a data-efficient way for training deep learning models. One-shot subset selection poses additional challenges as subset selection is only performed once and full set data become unavailable after the selection. However, most existing methods tend to choose either diverse or difficult data samples, which fail to faithfully represent the joint data distribution that is comprised of both feature and label information. The selection is also performed independently from the subset size, which plays an essential role in choosing what types of samples. To address this critical gap, we propose to conduct Feature similarity and Label variability Balanced One-shot Subset Selection (BOSS), aiming to construct an optimal size-aware subset for data-efficient deep learning. We show that a novel balanced core-set loss bound theoretically justifies the need to simultaneously consider both diversity and difficulty to form an optimal subset. It also reveals how the subset size influences the bound. We further connect the inaccessible bound to a practical surrogate target which is tailored to subset sizes and varying levels of overall difficulty. We design a novel Beta-scoring importance function to delicately control the optimal balance of diversity and difficulty. Comprehensive experiments conducted on both synthetic and real data justify the important theoretical properties and demonstrate the superior performance of BOSS as compared with the competitive baselines.

Few-shot action recognition aims at quickly adapting a pre-trained model to the novel data with a distribution shift using only a limited number of samples. Key challenges include how to identify and leverage the transferable knowledge learned by the pre-trained model. We therefore propose CDTD, or Causal Domain-Invariant Temporal Dynamics for knowledge transfer. To identify the temporally invariant and variant representations, we employ the causal representation learning methods for unsupervised pertaining, and then tune the classifier with supervisions in next stage. Specifically, we assume the domain information can be well estimated and the pre-trained temporal dynamic generation and transition models can be well transferred. During adaptation, we fix the transferable temporal dynamics and update the image encoder and domain estimator. The efficacy of our approach is revealed by the superior accuracy of CDTD over leading alternatives across standard few-shot action recognition datasets.

Model-Agnostic Meta-Learning (MAML) and its variants have shown remarkable performance in scenarios characterized by a scarcity of labeled data during the training phase of machine learning models. Despite these successes, MAMLbased approaches encounter significant challenges when there is a substantial discrepancy in the distribution of training and testing tasks, resulting in inefficient learning and limited generalization across domains. Inspired by classical proportional-integral-derivative (PID) control theory, this study introduces a Layer-Adaptive PID (LA-PID) Optimizer, a MAML-based optimizer that employs efficient parameter optimization methods to dynamically adjust task-specific PID control gains at each layer of the network, conducting a first-principles analysis of optimal convergence conditions. A series of experiments conducted on four standard benchmark datasets demonstrate the efficacy of the LA-PID optimizer, indicating that LA-PID achieves state-oftheart performance in few-shot classification and cross-domain tasks, accomplishing these objectives with fewer training steps. Code is available on https://github.com/yuguopin/LA-PID.

We present Premier-TACO, a multitask feature representation learning approach designed to improve few-shot policy learning efficiency in sequential decision-making tasks. Premier-TACO leverages a subset of multitask offline datasets for pretraining a general feature representation, which captures critical environmental dynamics and is fine-tuned using minimal expert demonstrations. It advances the temporal action contrastive learning (TACO) objective, known for state-of-the-art results in visual control tasks, by incorporating a novel negative example sampling strategy. This strategy is crucial in significantly boosting TACO’s computational efficiency, making large-scale multitask offline pretraining feasible. Our extensive empirical evaluation in a diverse set of continuous control benchmarks including Deepmind Control Suite, MetaWorld, and LIBERO demonstrate Premier-TACO’s effective- ness in pretraining visual representations, significantly enhancing few-shot imitation learning of novel tasks.

Machine unlearning, the study of efficiently removing the impact of specific training instances on a model, has garnered increased attention in recent years due to regulatory guidelines such as the Right to be Forgotten. Achieving precise unlearning typically involves fully retraining the model and is computationally infeasible in case of very large models such as Large Language Models (LLMs). To this end, recent work has proposed several algorithms which approximate the removal of training data without retraining the model. These algorithms crucially rely on access to the model parameters in order to update them, an assumption that may not hold in practice due to computational constraints or having only query access to the LLMs. In this work, we propose a new class of unlearning methods for LLMs called ``In-Context Unlearning.'' This method unlearns instances from the model by simply providing specific kinds of inputs in context, without the need to update model parameters. To unlearn specific training instances, we present these instances to the LLMs at inference time along with labels that differ from their ground truth. Our experimental results demonstrate that in-context unlearning performs on par with, or in some cases outperforms other state-of-the-art methods that require access to model parameters, effectively removing the influence of specific instances on the model while preserving test accuracy.

We demonstrate that large language models (LLMs) may learn indicators of document usefulness and modulate their updates accordingly. We introduce random strings ("tags") as indicators of usefulness in a synthetic fine-tuning dataset. Fine-tuning on this dataset leads to **implicit meta-learning (IML)**: in further fine-tuning, the model updates to make more use of text that is tagged as useful. We perform a thorough empirical investigation of this phenomenon, finding (among other things) that (i) it occurs in both pretrained LLMs and those trained from scratch, as well as on a vision task, and (ii) larger models and smaller batch sizes tend to give more IML. We also use probing to examine how IML changes the way models store knowledge in their parameters. Finally, we reflect on what our results might imply about the capabilities, risks, and controllability of future AI systems.

Few-shot open-set recognition (FSOSR) aims to detect instances from unseen classes by utilizing a small set of labeled instances from closed-set classes. Accurately rejecting instances from open-set classes in the few-shot setting is fundamentally more challenging due to the weaker supervised signals resulting from fewer labels. Transformer-based few-shot methods exploit attention mapping to achieve a consistent representation. However, the softmax-generated attention map normalizes all the instances that assign unnecessary high attentive weights to those instances not close to the closed-set classes that negatively impact the detection performance. In addition, open-set samples that are similar to a certain closed-set class also pose a significant challenge to most existing FSOSR models. To address these challenges, we propose a novel Meta Evidential Transformer (MET) based FSOSR model that uses an evidential open-set loss to learn more compact closed-set class representations by effectively leveraging similar closed-set classes. MET further integrates an evidence-to-variance ratio to detect fundamentally challenging tasks and uses an evidence-guided cross-attention mechanism to better separate the difficult open-set samples. Experiments on real-world datasets demonstrate consistent improvement over existing competitive methods in unseen class recognition without deteriorating closed-set performance.

In the Generalized Mastermind problem, there is an unknown subset $H$ of the hypercube 0,1$^d$ containing $n$ points. The goal is to learn $H$ by making a few queries to an oracle which given a point $q$ in 0,1$^d$, returns the point in $H$ nearest to $q$. We give a two-round adaptive algorithm for this problem that learns $H$ while making at most $\exp(\widetilde{O}(\sqrt{d \log n}))$. Furthermore, we show that any $r$-round adaptive randomized algorithm that learns $H$ with constant probability must make $\exp(\Omega(d^{3^{-(r-1)}}))$ queries even when the input has poly$(d)$ points; thus, any poly$(d)$ query algorithm must necessarily use $\Omega(\log \log d)$ rounds of adaptivity. We give optimal query complexity bounds for the variant of the problem where queries are allowed to be from 0,1,2$^d$. We also study a continuous variant of the problem in which $H$ is a subset of unit vectors in $\mathbb{R}^d$ and one can query unit vectors in $\mathbb{R}^d$. For this setting, we give a $O(n^{\lfloor d/2 \rfloor})$ query deterministic algorithm to learn the hidden set of points.

Bayesian few-shot classification has been a focal point in the field of few-shot learning. This paper seamlessly integrates mirror descent-based variational inference into Gaussian process-based few-shot classification, addressing the challenge of non-conjugate inference. By leveraging non-Euclidean geometry, mirror descent achieves accelerated convergence by providing the steepest descent direction along the corresponding manifold. It also exhibits the parameterization invariance property concerning the variational distribution. Experimental results demonstrate competitive classification accuracy, improved uncertainty quantification, and faster convergence compared to baseline models. Additionally, we investigate the impact of hyperparameters and components. Code is publicly available at https://github.com/keanson/MD-BSFC.

Denoising Diffusion Probabilistic Models (DDPMs) exhibit remarkable capabilities in image generation, with studies suggesting that they can generalize by composing latent factors learned from the training data. In this work, we go further and study DDPMs trained on strictly separate subsets of the data distribution with large gaps on the support of the latent factors. We show that such a model can effectively generate images in the unexplored, intermediate regions of the distribution. For instance, when trained on clearly smiling and non-smiling faces, we demonstrate a sampling procedure which can generate slightly smiling faces without reference images (zero-shot interpolation). We replicate these findings for other attributes as well as other datasets. [Our code is available on GitHub](https://github.com/jdeschena/ddpm-zero-shot-interpolation).

The few-shot fine-tuning of Latent Diffusion Models (LDMs) has enabled them to grasp new concepts from a limited number of images. However, given the vast amount of personal images accessible online, this capability raises critical concerns about civil privacy. While several previous defense methods have been developed to prevent such misuse of LDMs, they typically assume that the textual prompts used by data protectors exactly match those employed by data exploiters. In this paper, we first empirically demonstrate that breaking this assumption, i.e., in cases where discrepancies exist between the textual conditions used by protectors and exploiters, could substantially reduces the effectiveness of these defenses. Furthermore, considering the visual encoder's independence from textual prompts, we delve into the visual encoder and thoroughly investigate how manipulating the visual encoder affects the few-shot fine-tuning process of LDMs. Drawing on these insights, we propose a simple yet effective method called Prompt-Independent Defense (PID) to safeguard privacy against LDMs. We show that PID can act as a strong privacy shield on its own while requiring significantly less computational power. We believe our studies, along with the comprehensive understanding and new defense method, provide a notable advance toward reliable data protection against LDMs.

Pre-trained vision transformers have revolutionized few-shot image classification, and it has been recently demonstrated that the previous common practice of meta-learning in synergy with these pre-trained transformers still holds significance. In this work, we design a new framework centered exclusively on self-attention, called MetaFormer, which extends the vision transformers beyond patch token interactions to encompass relationships between samples and tasks simultaneously for further advancing their downstream task performance. Leveraging the intrinsical property of ViTs in handling local patch relationships, we propose Masked Sample Attention (MSA) to efficiently embed the sample relationships into the network, where an adaptive mask is attached for enhancing task-specific feature consistency and providing flexibility in switching between few-shot learning setups. To encapsulate task relationships while filtering out background noise, Patch-grained Task Attention (PTA) is designed to maintain a dynamic knowledge pool consolidating diverse patterns from historical tasks. MetaFormer demonstrates coherence and compatibility with off-the-shelf pre-trained vision transformers and shows significant improvements in both inductive and transductive few-shot learning scenarios, outperforming state-of-the-art methods by up to 8.77% and 6.25% on 12 in-domain and 10 cross-domain datasets, respectively.