Long Context

To make reinforcement learning more sample efficient, we need better credit assignment methods that measure an action’s influence on future rewards. Building upon Hindsight Credit Assignment (HCA), we introduce Counterfactual Contribution Analysis (COCOA), a new family of model-based credit assignment algorithms. Our algorithms achieve precise credit assignment by measuring the contribution of actions upon obtaining subsequent rewards, by quantifying a counterfactual query: ‘Would the agent still have reached this reward if it had taken another action?’. We show that measuring contributions w.r.t. rewarding _states_, as is done in HCA, results in spurious estimates of contributions, causing HCA to degrade towards the high-variance REINFORCE estimator in many relevant environments. Instead, we measure contributions w.r.t. rewards or learned representations of the rewarding objects, resulting in gradient estimates with lower variance. We run experiments on a suite of problems specifically designed to evaluate long-term credit assignment capabilities. By using dynamic programming, we measure ground-truth policy gradients and show that the improved performance of our new model-based credit assignment methods is due to lower bias and variance compared to HCA and common baselines. Our results demonstrate how modeling action contributions towards rewarding outcomes can be leveraged for credit assignment, opening a new path towards sample-efficient reinforcement learning.

Genomic (DNA) sequences encode an enormous amount of information for gene regulation and protein synthesis. Similar to natural language models, researchers have proposed foundation models in genomics to learn generalizable features from unlabeled genome data that can then be fine-tuned for downstream tasks such as identifying regulatory elements. Due to the quadratic scaling of attention, previous Transformer-based genomic models have used 512 to 4k tokens as context (<0.001% of the human genome), significantly limiting the modeling of long-range interactions in DNA. In addition, these methods rely on tokenizers or fixed k-mers to aggregate meaningful DNA units, losing single nucleotide resolution (i.e. DNA "characters") where subtle genetic variations can completely alter protein function via single nucleotide polymorphisms (SNPs). Recently, Hyena, a large language model based on implicit convolutions was shown to match attention in quality while allowing longer context lengths and lower time complexity. Leveraging Hyena’s new long-range capabilities, we present HyenaDNA, a genomic foundation model pretrained on the human reference genome with context lengths of up to 1 million tokens at the single nucleotide-level – an up to 500x increase over previous dense attention-based models. HyenaDNA scales sub-quadratically in sequence length (training up to 160x faster than Transformer), uses single nucleotide tokens, and has full global context at each layer. We explore what longer context enables - including the first use of in-context learning in genomics for simple adaptation to novel tasks without updating pretrained model weights. On fine-tuned benchmarks from the Nucleotide Transformer, HyenaDNA reaches state-of-the-art (SotA) on 12 of 18 datasets using a model with orders of magnitude less parameters and pretraining data.1 On the GenomicBenchmarks, HyenaDNA surpasses SotA on 7 of 8 datasets on average by +10 accuracy points. Code at https://github.com/HazyResearch/hyena-dna.

Large language models have shown tremendous performance in a variety of tasks. In-context learning -- the ability to improve at a task after being provided with a number of demonstrations -- is seen as one of the main contributors to their success. In the present paper, we demonstrate that the in-context learning abilities of large language models can be recursively improved via in-context learning itself. We coin this phenomenon meta-in-context learning. Looking at two idealized domains, a one-dimensional regression task and a two-armed bandit task, we show that meta-in-context learning adaptively reshapes a large language model's priors over expected tasks. Furthermore, we find that meta-in-context learning modifies the in-context learning strategies of such models. Finally, we broaden the scope of our investigation to encompass two diverse benchmarks: one focusing on real-world regression problems and the other encompassing multiple NLP tasks. In both cases, we observe competitive performance comparable to that of traditional learning algorithms. Taken together, our work improves our understanding of in-context learning and paves the way toward adapting large language models to the environment they are applied purely through meta-in-context learning rather than traditional finetuning.

We propose SutraNets, a novel method for neural probabilistic forecasting of long-sequence time series. SutraNets use an autoregressive generative model to factorize the likelihood of long sequences into products of conditional probabilities. When generating long sequences, most autoregressive approaches suffer from harmful error accumulation, as well as challenges in modeling long-distance dependencies. SutraNets treat long, univariate prediction as multivariate prediction over lower-frequency sub-series. Autoregression proceeds across time and across sub-series in order to ensure coherent multivariate (and, hence, high-frequency univariate) outputs. Since sub-series can be generated using fewer steps, SutraNets effectively reduce error accumulation and signal path distances. We find SutraNets to significantly improve forecasting accuracy over competitive alternatives on six real-world datasets, including when we vary the number of sub-series and scale up the depth and width of the underlying sequence models.

Large language models have an exceptional capability to incorporate new information in a contextual manner. However, the full potential of such an approach is often restrained due to a limitation in the effective context length. One solution to this issue is to endow an attention layer with access to an additional context, which comprises of (key, value) pairs. Yet, as the number of documents increases, the proportion of relevant keys to irrelevant ones decreases, leading the model to focus more on the irrelevant keys. We identify a significant challenge, dubbed the distraction issue, where keys linked to different semantic values might overlap, making them hard to distinguish. To tackle this problem, we introduce the Focused Transformer (FoT), a technique that employs a training process inspired by contrastive learning. This novel approach enhances the structure of the (key, value) space, enabling an extension of the context length. Our method allows for fine-tuning pre-existing, large-scale models to lengthen their effective context. This is demonstrated by our fine-tuning of $3 B$ and $7 B$ OpenLLaMA checkpoints. The resulting models, which we name LongLLaMA, exhibit advancements in tasks requiring a long context. We further illustrate that our LongLLaMA models adeptly manage a $256 k$ context length for passkey retrieval.

We consider the class of noisy multi-layered sigmoid recurrent neural networks with $w$ (unbounded) weights for classification of sequences of length $T$, where independent noise distributed according to $\mathcal{N}(0,\sigma^2)$ is added to the output of each neuron in the network. Our main result shows that the sample complexity of PAC learning this class can be bounded by $O (w\log(T/\sigma))$. For the non-noisy version of the same class (i.e., $\sigma=0$), we prove a lower bound of $\Omega (wT)$ for the sample complexity. Our results indicate an exponential gap in the dependence of sample complexity on $T$ for noisy versus non-noisy networks. Moreover, given the mild logarithmic dependence of the upper bound on $1/\sigma$, this gap still holds even for numerically negligible values of $\sigma$.

In-context learning is the ability of a pretrained model to adapt to novel and diverse downstream tasks by conditioning on prompt examples, without optimizing any parameters. While large language models have demonstrated this ability, how in-context learning could be performed over graphs is unexplored. In this paper, we develop \textbf{Pr}etraining \textbf{O}ver \textbf{D}iverse \textbf{I}n-Context \textbf{G}raph S\textbf{y}stems (PRODIGY), the first pretraining framework that enables in-context learning over graphs. The key idea of our framework is to formulate in-context learning over graphs with a novel \emph{prompt graph} representation, which connects prompt examples and queries. We then propose a graph neural network architecture over the prompt graph and a corresponding family of in-context pretraining objectives. With PRODIGY, the pretrained model can directly perform novel downstream classification tasks on unseen graphs via in-context learning. We provide empirical evidence of the effectiveness of our framework by showcasing its strong in-context learning performance on tasks involving citation networks and knowledge graphs. Our approach outperforms the in-context learning accuracy of contrastive pretraining baselines with hard-coded adaptation by 18\% on average across all setups. Moreover, it also outperforms standard finetuning with limited data by 33\% on average with in-context learning.

With the rise of large-scale models trained on broad data, in-context learning has become a new learning paradigm that has demonstrated significant potential in natural language processing and computer vision tasks. Meanwhile, in-context learning is still largely unexplored in the 3D point cloud domain. Although masked modeling has been successfully applied for in-context learning in 2D vision, directly extending it to 3D point clouds remains a formidable challenge. In the case of point clouds, the tokens themselves are the point cloud positions (coordinates) that are masked during inference. Moreover, position embedding in previous works may inadvertently introduce information leakage. To address these challenges, we introduce a novel framework, named Point-In-Context, designed especially for in-context learning in 3D point clouds, where both inputs and outputs are modeled as coordinates for each task. Additionally, we propose the Joint Sampling module, carefully designed to work in tandem with the general point sampling operator, effectively resolving the aforementioned technical issues. We conduct extensive experiments to validate the versatility and adaptability of our proposed methods in handling a wide range of tasks. Furthermore, with a more effective prompt selection strategy, our framework surpasses the results of individually trained models.

Binary Balanced Tree Recursive Neural Networks (BBT-RvNNs) enforce sequence composition according to a preset balanced binary tree structure. Thus, their non-linear recursion depth (which is the tree depth) is just $\log_2 n$ ($n$ being the sequence length). Such logarithmic scaling makes BBT-RvNNs efficient and scalable on long sequence tasks such as Long Range Arena (LRA). However, such computational efficiency comes at a cost because BBT-RvNNs cannot solve simple arithmetic tasks like ListOps. On the flip side, RvNN models (e.g., Beam Tree RvNN) that do succeed on ListOps (and other structure-sensitive tasks like formal logical inference) are generally several times more expensive (in time and space) than even Recurrent Neural Networks. In this paper, we introduce a novel framework --- Recursion in Recursion (RIR) to strike a balance between the two sides - getting some of the benefits from both worlds. In RIR, we use a form of two-level nested recursion - where the outer recursion is a $k$-ary balanced tree model with another recursive model (inner recursion) implementing its cell function. For the inner recursion, we choose Beam Tree RvNNs. To adjust Beam Tree RvNNs within RIR we also propose a novel strategy of beam alignment. Overall, this entails that the total recursive depth in RIR is upper-bounded by $k \log_k n$. Our best RIR-based model is the first model that demonstrates high ($\geq 90\%$) length-generalization performance on ListOps while at the same time being scalable enough to be trainable on long sequence inputs from LRA (it can reduce the memory usage of the original Beam Tree RvNN by hundreds of times). Moreover, in terms of accuracy in the LRA language tasks, it performs competitively with Structured State Space Models (SSMs) without any special initialization - outperforming Transformers by a large margin. On the other hand, while SSMs can marginally outperform RIR on LRA, they (SSMs) fail to length-generalize on ListOps. Our code is available at: https://github.com/JRC1995/BeamRecursionFamily/

Neural sequence models based on the transformer architecture have demonstrated remarkable \emph{in-context learning} (ICL) abilities, where they can perform new tasks when prompted with training and test examples, without any parameter update to the model. This work first provides a comprehensive statistical theory for transformers to perform ICL. Concretely, we show that transformers can implement a broad class of standard machine learning algorithms in context, such as least squares, ridge regression, Lasso, learning generalized linear models, and gradient descent on two-layer neural networks, with near-optimal predictive power on various in-context data distributions. Using an efficient implementation of in-context gradient descent as the underlying mechanism, our transformer constructions admit mild size bounds, and can be learned with polynomially many pretraining sequences. Building on these ``base'' ICL algorithms, intriguingly, we show that transformers can implement more complex ICL procedures involving \emph{in-context algorithm selection}, akin to what a statistician can do in real life---A \emph{single} transformer can adaptively select different base ICL algorithms---or even perform qualitatively different tasks---on different input sequences, without any explicit prompting of the right algorithm or task. We both establish this in theory by explicit constructions, and also observe this phenomenon experimentally. In theory, we construct two general mechanisms for algorithm selection with concrete examples: pre-ICL testing, and post-ICL validation. As an example, we use the post-ICL validation mechanism to construct a transformer that can perform nearly Bayes-optimal ICL on a challenging task---noisy linear models with mixed noise levels. Experimentally, we demonstrate the strong in-context algorithm selection capabilities of standard transformer architectures.

The dominant paradigm in 3D human pose estimation that lifts a 2D pose sequence to 3D heavily relies on long-term temporal clues (i.e., using a daunting number of video frames) for improved accuracy, which incurs performance saturation, intractable computation and the non-causal problem. This can be attributed to their inherent inability to perceive spatial context as plain 2D joint coordinates carry no visual cues. To address this issue, we propose a straightforward yet powerful solution: leveraging the $\textit{readily available}$ intermediate visual representations produced by off-the-shelf (pre-trained) 2D pose detectors -- no finetuning on the 3D task is even needed. The key observation is that, while the pose detector learns to localize 2D joints, such representations (e.g., feature maps) implicitly encode the joint-centric spatial context thanks to the regional operations in backbone networks. We design a simple baseline named $\textbf{Context-Aware PoseFormer}$ to showcase its effectiveness. $\textit{Without access to any temporal information}$, the proposed method significantly outperforms its context-agnostic counterpart, PoseFormer, and other state-of-the-art methods using up to $\textit{hundreds of}$ video frames regarding both speed and precision. $\textit{Project page:}$ https://qitaozhao.github.io/ContextAware-PoseFormer

Since the proposal of transformers, these models have been limited to bounded input lengths, because of their need to attend to every token in the input. In this work, we propose Unlimiformer: a general approach that wraps any existing pretrained encoder-decoder transformer, and offloads the cross-attention computation to a single $k$-nearest-neighbor ($k$NN) index, while the returned $k$NN distances are the attention dot-product scores. This $k$NN index can be kept on either the GPU or CPU memory and queried in sub-linear time; this way, we can index practically unlimited input sequences, while every attention head in every decoder layer retrieves its top-$k$ keys, instead of attending to every key. We evaluate Unlimiformer on several long-document and book-summarization benchmarks, showing that it can process even **500k** token-long inputs from the BookSum dataset, without any input truncation at test time. We demonstrate that Unlimiformer improves pretrained models such as BART and Longformer by extending them to unlimited inputs without additional learned weights and without modifying their code. Our code and models are publicly available at https://github.com/abertsch72/unlimiformer , and support LLaMA-2 as well.

We tackle the problem of Persistent Independent Particles (PIPs), also called Tracking Any Point (TAP), in videos, which specifically aims at estimating persistent long-term trajectories of query points in videos. Previous methods attempted to estimate these trajectories independently to incorporate longer image sequences, therefore, ignoring the potential benefits of incorporating spatial context features. We argue that independent video point tracking also demands spatial context features. To this end, we propose a novel framework Context-PIPs, which effectively improves point trajectory accuracy by aggregating spatial context features in videos. Context-PIPs contains two main modules: 1) a SOurse Feature Enhancement (SOFE) module, and 2) a TArget Feature Aggregation (TAFA) module. Context-PIPs significantly improves PIPs all-sided, reducing 11.4\% Average Trajectory Error of Occluded Points (ATE-Occ) on CroHD and increasing 11.8\% Average Percentage of Correct Keypoint (A-PCK) on TAP-Vid-Kinetics. Demos are available at \url{https://wkbian.github.io/Projects/Context-PIPs/}.

Time-dependent partial differential equations (PDEs) are ubiquitous in science and engineering. Recently, mostly due to the high computational cost of traditional solution techniques, deep neural network based surrogates have gained increased interest. The practical utility of such neural PDE solvers relies on their ability to provide accurate, stable predictions over long time horizons, which is a notoriously hard problem. In this work, we present a large-scale analysis of common temporal rollout strategies, identifying the neglect of non-dominant spatial frequency information, often associated with high frequencies in PDE solutions, as the primary pitfall limiting stable, accurate rollout performance. Based on these insights, we draw inspiration from recent advances in diffusion models to introduce PDE-Refiner; a novel model class that enables more accurate modeling of all frequency components via a multistep refinement process. We validate PDE-Refiner on challenging benchmarks of complex fluid dynamics, demonstrating stable and accurate rollouts that consistently outperform state-of-the-art models, including neural, numerical, and hybrid neural-numerical architectures. We further demonstrate that PDE-Refiner greatly enhances data efficiency, since the denoising objective implicitly induces a novel form of spectral data augmentation. Finally, PDE-Refiner's connection to diffusion models enables an accurate and efficient assessment of the model's predictive uncertainty, allowing us to estimate when the surrogate becomes inaccurate.

Transformers have emerged as the cornerstone of state-of-the-art natural language processing models, showcasing exceptional performance across a wide range of AI applications. However, the memory demands posed by the self-attention mechanism and the large feedforward network in Transformers limit their ability to handle long sequences, thereby creating challenges for tasks involving multiple long sequences or long-term dependencies. We present a distinct approach, Blockwise Parallel Transformer (BPT), that leverages blockwise computation of self-attention and feedforward network fusion to minimize memory costs. By processing longer input sequences while maintaining memory efficiency, BPT enables training sequences 32 times longer than vanilla Transformers and up to 4 times longer than previous memory-efficient methods. Extensive experiments on language modeling and reinforcement learning tasks demonstrate the effectiveness of BPT in reducing memory requirements and improving performance.

The discovery of neural architectures from simple building blocks is a long-standing goal of Neural Architecture Search (NAS). Hierarchical search spaces are a promising step towards this goal but lack a unifying search space design framework and typically only search over some limited aspect of architectures. In this work, we introduce a unifying search space design framework based on context-free grammars that can naturally and compactly generate expressive hierarchical search spaces that are 100s of orders of magnitude larger than common spaces from the literature. By enhancing and using their properties, we effectively enable search over the complete architecture and can foster regularity. Further, we propose an efficient hierarchical kernel design for a Bayesian Optimization search strategy to efficiently search over such huge spaces. We demonstrate the versatility of our search space design framework and show that our search strategy can be superior to existing NAS approaches. Code is available at https://github.com/automl/hierarchical_nas_construction.

We consider a contextual bandit problem with $S $ contexts and $K $ actions. In each round $t=1,2,\dots$ the learner observes a random context and chooses an action based on its past experience. The learner then observes a random reward whose mean is a function of the context and the action for the round. Under the assumption that the contexts can be lumped into $r\le \min(S ,K)$ groups such that the mean reward for the various actions is the same for any two contexts that are in the same group, we give an algorithm that outputs an $\epsilon$-optimal policy after using at most $\widetilde O(r (S +K )/\epsilon^2)$ samples with high probability and provide a matching $\widetilde\Omega(r (S +K )/\epsilon^2)$ lower bound. In the regret minimization setting, we give an algorithm whose cumulative regret up to time $T$ is bounded by $\widetilde O(\sqrt{r ^3(S +K )T})$. To the best of our knowledge, we are the first to show the near-optimal sample complexity in the PAC setting and $\widetilde O{\sqrt{\text{poly}(r)(S+K)T}}$ minimax regret in the online setting for this problem. We also show our algorithms can be applied to more general low-rank bandits and get improved regret bounds in some scenarios.

This paper presents the official release of the Digital Typhoon dataset, the longest typhoon satellite image dataset for 40+ years aimed at benchmarking machine learning models for long-term spatio-temporal data. To build the dataset, we developed a workflow to create an infrared typhoon-centered image for cropping using Lambert azimuthal equal-area projection referring to the best track data. We also address data quality issues such as inter-satellite calibration to create a homogeneous dataset. To take advantage of the dataset, we organized machine learning tasks by the types and targets of inference, with other tasks for meteorological analysis, societal impact, and climate change. The benchmarking results on the analysis, forecasting, and reanalysis for the intensity suggest that the dataset is challenging for recent deep learning models, due to many choices that affect the performance of various models. This dataset reduces the barrier for machine learning researchers to meet large-scale real-world events called tropical cyclones and develop machine learning models that may contribute to advancing scientific knowledge on tropical cyclones as well as solving societal and sustainability issues such as disaster reduction and climate change. The dataset is publicly available at http://agora.ex.nii.ac.jp/digital-typhoon/dataset/ and https://github.com/kitamoto-lab/digital-typhoon/.