NeurIPS 2024 Papers

Neural networks are often trained on multiple tasks, either simultaneously (multi-task learning, MTL) or sequentially (pretraining and subsequent finetuning, PT+FT). In particular, it is common practice to pretrain neural networks on a large auxiliary task before finetuning on a downstream task with fewer samples. Despite the prevalence of this approach, the inductive biases that arise from learning multiple tasks are poorly characterized. In this work, we address this gap. We describe novel implicit regularization penalties associated with MTL and PT+FT in diagonal linear networks and single-hidden-layer ReLU networks. These penalties indicate that MTL and PT+FT induce the network to reuse features in different ways. 1) Both MTL and PT+FT exhibit biases towards feature reuse between tasks, and towards sparsity in the set of learned features. We show a "conservation law" that implies a direct tradeoff between these two biases. 2) PT+FT exhibits a novel "nested feature selection" regime, not described by either the "lazy" or "rich" regimes identified in prior work, which biases it to *rely on a sparse subset* of the features learned during pretraining. This regime is much narrower for MTL. 3) PT+FT (but not MTL) in ReLU networks benefits from features that are correlated between the auxiliary and main task. We confirm these findings empirically with teacher-student models, and introduce a technique -- weight rescaling following pretraining -- that can elicit the nested feature selection regime. Finally, we validate our theory in deep neural networks trained on image classification. We find that weight rescaling improves performance when it causes models to display signatures of nested feature selection. Our results suggest that nested feature selection may be an important inductive bias for finetuning neural networks.

Hair editing is a critical image synthesis task that aims to edit hair color and hairstyle using text descriptions or reference images, while preserving irrelevant attributes (e.g., identity, background, cloth). Many existing methods are based on StyleGAN to address this task. However, due to the limited spatial distribution of StyleGAN, it struggles with multiple hair color editing and facial preservation. Considering the advancements in diffusion models, we utilize Latent Diffusion Models (LDMs) for hairstyle editing. Our approach introduces Multi-stage Hairstyle Blend (MHB), effectively separating control of hair color and hairstyle in diffusion latent space. Additionally, we train a warping module to align the hair color with the target region. To further enhance multi-color hairstyle editing, we fine-tuned a CLIP model using a multi-color hairstyle dataset. Our method not only tackles the complexity of multi-color hairstyles but also addresses the challenge of preserving original colors during diffusion editing. Extensive experiments showcase the superiority of our method in editing multi-color hairstyles while preserving facial attributes given textual descriptions and reference images.

Motivated by the strategic participation of electricity producers in electricity day-ahead market, we study the problem of online learning in repeated multi-unit uniform price auctions focusing on the adversarial opposing bid setting. The main contribution of this paper is the introduction of a new modeling of the bid space. Indeed, we prove that a learning algorithm leveraging the structure of this problem achieves a regret of $\tilde{O}(K^{4/3}T^{2/3})$ under bandit feedback, improving over the bound of $\tilde{O}(K^{7/4}T^{3/4})$ previously obtained in the literature. This improved regret rate is tight up to logarithmic terms. %by deducing a lower bound of $\Omega (T^{2/3})$ from the dynamic pricing literature, proving the optimality in $T$ of our algorithm up to log factors. Inspired by electricity reserve markets, we further introduce a different feedback model under which all winning bids are revealed. This feedback interpolates between the full-information and bandit scenarios depending on the auctions' results. We prove that, under this feedback, the algorithm that we propose achieves regret $\tilde{O}(K^{5/2}\sqrt{T})$.

Recently, rapid advancements in Multi-Modal In-Context Learning (MM-ICL) have achieved notable success, which is capable of achieving superior performance across various tasks without requiring additional parameter tuning. However, the underlying rules for the effectiveness of MM-ICL remain under-explored. To fill this gap, this work aims to investigate the research question: "_What factors affect the performance of MM-ICL?_" To this end, we investigate extensive experiments on the three core steps of MM-ICL including demonstration retrieval, demonstration ordering, and prompt construction using 6 vision large language models and 20 strategies. Our findings highlight (1) the necessity of a multi-modal retriever for demonstration retrieval, (2) the importance of intra-demonstration ordering over inter-demonstration ordering, and (3) the enhancement of task comprehension through introductory instructions in prompts. We hope this study can serve as a foundational guide for optimizing MM-ICL strategies in future research.

In recent years, significant progress has been made in multi-objective reinforcement learning (RL) research, which aims to balance multiple objectives by incorporating preferences for each objective. In most existing studies, specific preferences must be provided during deployment to indicate the desired policies explicitly. However, designing these preferences depends heavily on human prior knowledge, which is typically obtained through extensive observation of high-performing demonstrations with expected behaviors. In this work, we propose a simple yet effective offline adaptation framework for multi-objective RL problems without assuming handcrafted target preferences, but only given several demonstrations to implicitly indicate the preferences of expected policies. Additionally, we demonstrate that our framework can naturally be extended to meet constraints on safety-critical objectives by utilizing safe demonstrations, even when the safety thresholds are unknown. Empirical results on offline multi-objective and safe tasks demonstrate the capability of our framework to infer policies that align with real preferences while meeting the constraints implied by the provided demonstrations.

We propose a novel piecewise stationary linear bandit (PSLB) model, where the environment randomly samples a context from an unknown probability distribution at each changepoint, and the quality of an arm is measured by its return averaged over all contexts. The contexts and their distribution, as well as the changepoints are unknown to the agent. We design Piecewise-Stationary $\varepsilon$-Best Arm Identification$^+$ (PS$\varepsilon$BAI$^+$), an algorithm that is guaranteed to identify an $\varepsilon$-optimal arm with probability $\ge 1-\delta$ and with a minimal number of samples. PS$\varepsilon$BAI$^+$ consists of two subroutines, PS$\varepsilon$BAI and Naïve $\varepsilon$-BAI (N$\varepsilon$BAI), which are executed in parallel. PS$\varepsilon$BAI actively detects changepoints and aligns contexts to facilitate the arm identification process. When PS$\varepsilon$BAI and N$\varepsilon$BAI are utilized judiciously in parallel, PS$\varepsilon$BAI$^+$ is shown to have a finite expected sample complexity. By proving a lower bound, we show the expected sample complexity of PS$\varepsilon$BAI$^+$ is optimal up to a logarithmic factor. We compare PS$\varepsilon$BAI$^+$ to baseline algorithms using numerical experiments which demonstrate its efficiency. Both our analytical and numerical results corroborate that the efficacy of PS$\varepsilon$BAI$^+$ is due to the delicate change detection and context alignment procedures embedded in PS$\varepsilon$BAI.

Language models trained on diverse datasets unlock generalization by in-context learning. Reinforcement Learning (RL) policies can achieve a similar effect by meta-learning within the memory of a sequence model. However, meta-RL research primarily focuses on adapting to minor variations of a single task. It is difficult to scale towards more general behavior without confronting challenges in multi-task optimization, and few solutions are compatible with meta-RL's goal of learning from large training sets of unlabeled tasks. To address this challenge, we revisit the idea that multi-task RL is bottlenecked by imbalanced training losses created by uneven return scales across different tasks. We build upon recent advancements in Transformer-based (in-context) meta-RL and evaluate a simple yet scalable solution where both an agent's actor and critic objectives are converted to classification terms that decouple optimization from the current scale of returns. Large-scale comparisons in Meta-World ML45, Multi-Game Procgen, Multi-Task POPGym, Multi-Game Atari, and BabyAI find that this design unlocks significant progress in online multi-task adaptation and memory problems without explicit task labels.

Multi-instance partial-label learning (MIPL) is an emerging learning framework where each training sample is represented as a multi-instance bag associated with a candidate label set. Existing MIPL algorithms often overlook the margins for attention scores and predicted probabilities, leading to suboptimal generalization performance. A critical issue with these algorithms is that the highest prediction probability of the classifier may appear on a non-candidate label. In this paper, we propose an algorithm named MIPLMA, i.e., Multi-Instance Partial-Label learning with Margin Adjustment, which adjusts the margins for attention scores and predicted probabilities. We introduce a margin-aware attention mechanism to dynamically adjust the margins for attention scores and propose a margin distribution loss to constrain the margins between the predicted probabilities on candidate and non-candidate label sets. Experimental results demonstrate the superior performance of MIPLMA over existing MIPL algorithms, as well as other well-established multi-instance learning algorithms and partial-label learning algorithms.

The process of satisfying daily demands is a fundamental aspect of humans' daily lives. With the advancement of embodied AI, robots are increasingly capable of satisfying human demands. Demand-driven navigation (DDN) is a task in which an agent must locate an object to satisfy a specified demand instruction, such as "I am thirsty." The previous study typically assumes that each demand instruction requires only one object to be fulfilled and does not consider individual preferences. However, the realistic human demand may involve multiple objects. In this paper, we introduce the Multi-object Demand-driven Navigation (MO-DDN) benchmark, which addresses these nuanced aspects, including multi-object search and personal preferences, thus making the MO-DDN task more reflective of real-life scenarios compared to DDN. Building upon previous work, we employ the concept of ``attribute'' to tackle this new task. However, instead of solely relying on attribute features in an end-to-end manner like DDN, we propose a modular method that involves constructing a coarse-to-fine attribute-based exploration agent (C2FAgent). Our experimental results illustrate that this coarse-to-fine exploration strategy capitalizes on the advantages of attributes at various decision-making levels, resulting in superior performance compared to baseline methods. Code and video can be found at https://sites.google.com/view/moddn.

Learn-to-Defer is a paradigm that enables learning algorithms to work not in isolation but as a team with human experts. In this paradigm, we permit the system to defer a subset of its tasks to the expert. Although there are currently systems that follow this paradigm and are designed to optimize the accuracy of the final human-AI team, the general methodology for developing such systems under a set of constraints (e.g., algorithmic fairness, expert intervention budget, defer of anomaly, etc.) remains largely unexplored. In this paper, using a d-dimensional generalization to the fundamental lemma of Neyman and Pearson (d-GNP), we obtain the Bayes optimal solution for learn-to-defer systems under various constraints. Furthermore, we design a generalizable algorithm to estimate that solution and apply this algorithm to the COMPAS, Hatespeech, and ACSIncome datasets. Our algorithm shows improvements in terms of constraint violation over a set of learn-to-defer baselines and can control multiple constraint violations at once. The use of d-GNP is beyond learn-to-defer applications and can potentially obtain a solution to decision-making problems with a set of controlled expected performance measures.

As a major unsupervised learning method, clustering has received a lot of attention over multiple decades. The various clustering problems that have been studied intensively include, e.g., the $k$-means problem and the $k$-center problem. However, in applications, it is common that good clusterings should optimize multiple objectives (e.g., visualizing data on a map by clustering districts into areas that are both geographically compact but also homogeneous with respect to the data). We study combinations of different objectives, for example optimizing $k$-center and $k$-means simultaneously or optimizing $k$-center with respect to two different metrics. Usually these objectives are conflicting and cannot be optimized simultaneously, making it necessary to find trade-offs. We develop novel algorithms for computing the set of Pareto-optimal solutions (approximately) for various combinations of two objectives. Our algorithms achieve provable approximation guarantees and we demonstrate in several experiments that the (approximate) Pareto set contains good clusterings that cannot be found by considering one of the objectives separately.

In the realm of music AI, arranging rich and structured multi-track accompaniments from a simple lead sheet presents significant challenges. Such challenges include maintaining track cohesion, ensuring long-term coherence, and optimizing computational efficiency. In this paper, we introduce a novel system that leverages prior modelling over disentangled style factors to address these challenges. Our method presents a two-stage process: initially, a piano arrangement is derived from the lead sheet by retrieving piano texture styles; subsequently, a multi-track orchestration is generated by infusing orchestral function styles into the piano arrangement. Our key design is the use of vector quantization and a unique multi-stream Transformer to model the long-term flow of the orchestration style, which enables flexible, controllable, and structured music generation. Experiments show that by factorizing the arrangement task into interpretable sub-stages, our approach enhances generative capacity while improving efficiency. Additionally, our system supports a variety of music genres and provides style control at different composition hierarchies. We further show that our system achieves superior coherence, structure, and overall arrangement quality compared to existing baselines.

Deep learning based neural decoding from stereotactic electroencephalography (sEEG) would likely benefit from scaling up both dataset and model size. To achieve this, combining data across multiple subjects is crucial. However, in sEEG cohorts, each subject has a variable number of electrodes placed at distinct locations in their brain, solely based on clinical needs. Such heterogeneity in electrode number/placement poses a significant challenge for data integration, since there is no clear correspondence of the neural activity recorded at distinct sites between individuals. Here we introduce seegnificant: a training framework and architecture that can be used to decode behavior across subjects using sEEG data. We tokenize the neural activity within electrodes using convolutions and extract long-term temporal dependencies between tokens using self-attention in the time dimension. The 3D location of each electrode is then mixed with the tokens, followed by another self-attention in the electrode dimension to extract effective spatiotemporal neural representations. Subject-specific heads are then used for downstream decoding tasks. Using this approach, we construct a multi-subject model trained on the combined data from 21 subjects performing a behavioral task. We demonstrate that our model is able to decode the trial-wise response time of the subjects during the behavioral task solely from neural data. We also show that the neural representations learned by pretraining our model across individuals can be transferred in a few-shot manner to new subjects. This work introduces a scalable approach towards sEEG data integration for multi-subject model training, paving the way for cross-subject generalization for sEEG decoding.

Image feature extractors are rendered substantially more useful if different views of the same 3D location yield similar features while still being distinct from other locations. A feature extractor that achieves this goal even under significant viewpoint changes must recognise not just semantic categories in a scene, but also understand how different objects relate to each other in three dimensions. Existing work addresses this task by posing it as a patch retrieval problem, training the extracted features to facilitate retrieval of all image patches that project from the same 3D location. However, this approach uses a loss formulation that requires substantial memory and computation resources, limiting its applicability for large-scale training. We present a method for memory-efficient learning of location-consistent features that reformulates and approximates the smooth average precision objective. This novel loss function enables improvements in memory efficiency by three orders of magnitude, mitigating a key bottleneck of previous methods and allowing much larger models to be trained with the same computational resources. We showcase the improved location consistency of our trained feature extractor directly on a multi-view consistency task, as well as the downstream task of scene-stable panoptic segmentation, significantly outperforming previous state-of-the-art.

Multi-objective reinforcement learning (MORL) is an excellent framework for multi-objective sequential decision-making. MORL employs a utility function to aggregate multiple objectives into one that expresses a user's preferences. However, MORL still misses two crucial theoretical analyses of the properties of utility functions: (1) a characterisation of the utility functions for which an associated optimal policy exists, and (2) a characterisation of the types of preferences that can be expressed as utility functions. As a result, we formally characterise the families of preferences and utility functions that MORL should focus on: those for which an optimal policy is guaranteed to exist. We expect our theoretical results to promote the development of novel MORL algorithms that exploit our theoretical findings.

Deep neural networks (DNNs) have showcased their remarkable precision in approximating smooth functions. However, they suffer from the {\it spectral bias}, wherein DNNs typically exhibit a tendency to prioritize the learning of lower-frequency components of a function, struggling to effectively capture its high-frequency features. This paper is to address this issue. Notice that a function having only low frequency components may be well-represented by a shallow neural network (SNN), a network having only a few layers. By observing that composition of low frequency functions can effectively approximate a high-frequency function, we propose to learn a function containing high-frequency components by composing several SNNs, each of which learns certain low-frequency information from the given data. We implement the proposed idea by exploiting the multi-grade deep learning (MGDL) model, a recently introduced model that trains a DNN incrementally, grade by grade, a current grade learning from the residue of the previous grade only an SNN (with trainable parameters) composed with the SNNs (with fixed parameters) trained in the preceding grades as features. We apply MGDL to synthetic, manifold, colored images, and MNIST datasets, all characterized by presence of high-frequency features. Our study reveals that MGDL excels at representing functions containing high-frequency information. Specifically, the neural networks learned in each grade adeptly capture some low-frequency information, allowing their compositions with SNNs learned in the previous grades effectively representing the high-frequency features. Our experimental results underscore the efficacy of MGDL in addressing the spectral bias inherent in DNNs. By leveraging MGDL, we offer insights into overcoming spectral bias limitation of DNNs, thereby enhancing the performance and applicability of deep learning models in tasks requiring the representation of high-frequency information. This study confirms that the proposed method offers a promising solution to address the spectral bias of DNNs. The code is available on GitHub: \href{https://github.com/Ronglong-Fang/AddressingSpectralBiasviaMGDL}{\texttt{Addressing Spectral Bias via MGDL}}.

With the remarkable achievements of large language models (LLMs), the demand for fine-tuning and deploying LLMs in various downstream tasks has garnered widespread interest. Parameter-efficient fine-tuning techniques represented by LoRA and model quantization techniques represented by GPTQ and AWQ are of paramount significance. However, although these techniques have been widely adopted in single-task scenarios, research is scarce in multi-task scenarios. To be specific, we find that mainstream quantization methods would prevent the base LLM from being shared among tasks, so current LLM serving systems are infeasible to integrate LLM quantization with multiple LoRA adapters to achieve memory-efficient multi-task serving. Moreover, existing LLM serving systems lack support for dynamic task addition and overlook the workload differences among tasks, leading to inefficiencies in multi-task scenarios. This work proposes LoRA-Inlaid, an efficient multi-task LLM serving system. On the one hand, LoRA-Inlaid designs a flexible and efficient multi-task quantization algorithm (MLGPTQ) that facilitates the sharing of a single quantized model for multiple LoRA adapters, which significantly reduces the memory consumption for model deployment. Meanwhile, it supports adding LoRA adapters for new tasks on the fly, without sacrificing the stability of online services. On the other hand, LoRA-Inlaid develops a novel multi-task scheduling algorithm guided by output length prediction and grouping among different tasks, which effectively shrinks the memory consumption and avoids frequent switching of LoRA adapters. Empirical results verify that LoRA-Inlaid outperforms existing state-of-the-art LLM serving systems by up to 1.58 times in terms of throughput, 1.76 times in terms of average latency, 2 times in terms of job completion time, and 10 times in terms of SLO Attainment, while maintaining the same level of model quality.

Multi-task offline reinforcement learning aims to develop a unified policy for diverse tasks without requiring real-time interaction with the environment. Recent work explores sequence modeling, leveraging the scalability of the transformer architecture as a foundation for multi-task learning. Given the variations in task content and complexity, formulating policies becomes a challenging endeavor, requiring careful parameter sharing and adept management of conflicting gradients to extract rich cross-task knowledge from multiple tasks and transfer it to unseen tasks. In this paper, we propose the Decomposed Prompt Decision Transformer (DPDT) that adopts a two-stage paradigm to efficiently learn prompts for unseen tasks in a parameter-efficient manner. We incorporate parameters from pre-trained language models (PLMs) to initialize DPDT, thereby providing rich prior knowledge encoded in language models. During the decomposed prompt tuning phase, we learn both cross-task and task-specific prompts on training tasks to achieve prompt decomposition. In the test time adaptation phase, the cross-task prompt, serving as a good initialization, were further optimized on unseen tasks through test time adaptation, enhancing the model's performance on these tasks. Empirical evaluation on a series of Meta-RL benchmarks demonstrates the superiority of our approach. The project is available at https://github.com/ruthless-man/DPDT.

In recent years, machine learning has demonstrated impressive capability in handling molecular science tasks. To support various molecular properties at scale, machine learning models are trained in the multi-task learning paradigm. Nevertheless, data of different molecular properties are often not aligned: some quantities, e.g. equilibrium structure, demand more cost to compute than others, e.g. energy, so their data are often generated by cheaper computational methods at the cost of lower accuracy, which cannot be directly overcome through multi-task learning. Moreover, it is not straightforward to leverage abundant data of other tasks to benefit a particular task. To handle such data heterogeneity challenges, we exploit the specialty of molecular tasks that there are physical laws connecting them, and design consistency training approaches that allow different tasks to exchange information directly so as to improve one another. Particularly, we demonstrate that the more accurate energy data can improve the accuracy of structure prediction. We also find that consistency training can directly leverage force and off-equilibrium structure data to improve structure prediction, demonstrating a broad capability for integrating heterogeneous data.