Scaling large models with long sequences across applications like language generation, video generation and multimodal tasks requires efficient sequence parallelism. However, existing sequence parallelism methods all assume a single sequence dimension and fail to adapt to multi-dimensional transformer architectures that perform attention calculations across different dimensions. This paper introduces Dynamic Sequence Parallelism (DSP), a novel approach to enable efficient sequence parallelism for multi-dimensional transformer models. The key idea is to dynamically switch the parallelism dimension according to the current computation stage, leveraging the potential characteristics of multi-dimensional attention. This dynamic dimension switching allows sequence parallelism with minimal communication overhead compared to applying traditional single-dimension parallelism to multi-dimensional models. Experiments show DSP improves end-to-end throughput by 42.0% to 216.8% over prior sequence parallelism methods.
We present GenEFT: an effective theory framework for shedding light on the statics and dynamics of neural network generalization, and illustrate it with graph learning examples. We first investigate the generalization phase transition as data size increases, comparing experimental results with information-theory-based approximations. We find generalization in a Goldilocks zone where the decoder is neither too weak nor too powerful. We then introduce an effective theory for the dynamics of representation learning, where latent-space representations are modeled as interacting particles (repons), and find that it explains our experimentally observed phase transition between generalization and overfitting as encoder and decoder learning rates are scanned. This highlights the power of physics-inspired effective theories for bridging the gap between theoretical predictions and practice in machine learning.
We present MIPS, a novel method for program synthesis based on automated mechanistic interpretability of neural networks trained to perform the desired task, auto-distilling the learned algorithm into Python code. We test MIPS on a benchmark of 62 algorithmic tasks that can be learned by an RNN and find it highly complementary to GPT-4: MIPS solves 32 of them, including 13 that are not solved by GPT-4 (which also solves 30). MIPS uses an integer autoencoder to convert the RNN into a finite state machine, then applies Boolean or integer symbolic regression to capture the learned algorithm. As opposed to large language models, this program synthesis technique makes no use of (and is therefore not limited by) human training data such as algorithms and code from GitHub. We discuss opportunities and challenges for scaling up this approach to make machine-learned models more interpretable and trustworthy.
Neural scaling laws characterize how model performance improves as the model size scales up. Inspired by empirical observations, we introduce a resource model of neural scaling. A task is usually composite hence can be decomposed into many subtasks, which compete for resources (measured by the number of neurons allocated to subtasks). On toy problems, we empirically find that: (1) The loss of a subtask is inversely proportional to its allocated neurons. (2) When multiple subtasks are present in a composite task, the resources acquired by each subtask uniformly grow as models get larger, keeping the ratios of acquired resources constants. We hypothesize these findings to be generally true and build a model to predict neural scaling laws for general composite tasks, which successfully replicates the neural scaling law of Chinchilla models reported in arXiv:2203.15556. We believe that the notion of resource used in this paper will be a useful tool for characterizing and diagnosing neural networks.
Diffusion models are capable of impressive feats of image generation with uncommon juxtapositions such as astronauts riding horses on the moon with properly placed shadows. These outputs indicate the ability to perform compositional generalization, but how do the models do so? We perform controlled experiments on conditional DDPMs learning to generate 2D spherical Gaussian bumps centered at specified $x$- and $y$-positions. Our results show that the emergence of semantically meaningful latent representations is key to achieving high performance. En route to successful performance over learning, the model traverses three distinct phases of latent representations: (phase A) no latent structure, (phase B) a 2D manifold of disordered states, and (phase C) a 2D ordered manifold. Corresponding to each of these phases, we identify qualitatively different generation behaviors: 1) multiple bumps are generated, 2) one bump is generated but at inaccurate $x$ and $y$ locations, 3) a bump is generated at the correct $x$ and y location. Furthermore, we show that even under imbalanced datasets where features ($x$- versus $y$-positions) are represented with skewed frequencies, the learning process for $x$ and $y$ is coupled rather than factorized, demonstrating that simple vanilla-flavored diffusion models cannot learn efficient representations in which localization in $x$ and $y$ are factorized into separate 1D tasks. These findings suggest the need for future work to find inductive biases that will push generative models to discover and exploit factorizable independent structures in their inputs, which will be required to vault these models into more data-efficient regimes.
Large deep learning models have achieved impressive performance across a range of applications. However, their large memory requirements, including parameter memory and activation memory, have become a significant challenge for their practical serving. While existing methods mainly address parameter memory, the importance of activation memory has been overlooked. Especially for long input sequences, activation memory is expected to experience a significant exponential growth as the length of sequences increases. In this approach, we propose AutoChunk, an automatic and adaptive compiler system that efficiently reduces activation memory for long sequence inference by chunk strategies. The proposed system generates chunk plans by optimizing through multiple stages. In each stage, the chunk search pass explores all possible chunk candidates and the chunk selection pass identifies the optimal one. At runtime, AutoChunk employs code generation to automatically apply chunk strategies. The experiments demonstrate that AutoChunk can reduce over 80\% of activation memory while maintaining speed loss within 10%, extend max sequence length by 3.2x to 11.7x, and outperform state-of-the-art methods by a large margin.
This paper provides a novel parsimonious yet efficient design for zero-shot learning (ZSL), dubbed ParsNets, where we are interested in learning a composition of on-device friendly linear networks, each with orthogonality and low-rankness properties, to achieve equivalent or even better performance against existing deep models. Concretely, we first refactor the core module of ZSL, i.e., visual-semantics mapping function, into several base linear networks that correspond to diverse components of the semantic space, where the complex nonlinearity can be collapsed into simple local linearities. Then, to facilitate the generalization of local linearities, we construct a maximal margin geometry on the learned features by enforcing low-rank constraints on intra-class samples and high-rank constraints on inter-class samples, resulting in orthogonal subspaces for different classes and each subspace lies on a compact manifold. To enhance the model's adaptability and counterbalance over/under-fittings in ZSL, a set of sample-wise indicators is employed to select a sparse subset from these base linear networks to form a composite semantic predictor for each sample. Notably, maximal margin geometry can guarantee the diversity of features, and meanwhile, local linearities guarantee efficiency. Thus, our ParsNets can generalize better to unseen classes and can be deployed flexibly on resource-constrained devices. Theoretical explanations and extensive experiments are conducted to verify the effectiveness of the proposed method.
An essential goal in mechanistic interpretability to decode a network, i.e., to convert a neural network's raw weights to an interpretable algorithm. Given the difficulty of the decoding problem, progress has been made to understand the easier encoding problem, i.e., to convert an interpretable algorithm into network weights. Previous works focus on encoding existing algorithms into networks, which are interpretable by definition. However, focusing on encoding limits the possibility of discovering new algorithms that humans have never stumbled upon, but that are nevertheless interpretable. In this work, we explore the possibility of using hypernetworks to generate interpretable networks whose underlying algorithms are not yet known. The hypernetwork is carefully designed such that it can control network complexity, leading to a diverse family of interpretable algorithms ranked by their complexity. All of them are interpretable in hindsight, although some of them are less intuitive to humans, hence providing new insights regarding how to "think" like a neural network. For the task of computing L1 norms, hypernetworks find three algorithms: (a) the double-sided algorithm, (b) the convexity algorithm, (c) the pudding algorithm, although only the first algorithm was expected by the authors before experiments. We automatically classify these algorithms and analyze how these algorithmic phases develop during training, as well as how they are affected by complexity control. Furthermore, we show that a trained hypernetwork can correctly construct models for input dimensions not seen in training, demonstrating systematic generalization.
Recurrent neural networks (RNNs) trained on compositional tasks can exhibit functional modularity, in which neurons can be clustered by activity similarity and participation in shared computational subtasks. Unlike brains, these RNNs do not exhibit anatomical modularity, in which functional clustering is correlated with strong recurrent coupling and spatial localization of functional clusters. Contrasting with functional modularity, which can be ephemerally dependent on the input, anatomically modular networks form a robust substrate for solving the same subtasks in the future. To examine whether it is possible to grow brain-like anatomical modularity, we apply a recent machine learning method, brain-inspired modular training (BIMT), to a network being trained to solve a set of compositional cognitive tasks. We find that functional and anatomical clustering emerge together, such that functionally similar neurons also become spatially localized and interconnected. Moreover, compared to standard $L_1$ or no regularization settings, the model exhibits superior performance by optimally balancing task performance and network sparsity. In addition to achieving brain-like organization in RNNs, our findings also suggest that BIMT holds promise for applications in neuromorphic computing and enhancing the interpretability of neural network architectures.
We attribute grokking, the phenomenon where generalization is much delayed after memorization, to compression. To do so, we define linear mapping number (LMN) to measure network complexity, which is a generalized version of linear region number for ReLU networks. LMN can nicely characterize neural network compression before generalization. Although the $L_2$ norm has been a popular choice for characterizing model complexity, we argue in favor of LMN for a number of reasons: (1) LMN can be naturally interpreted as information/computation, while $L_2$ cannot. (2) In the compression phase, LMN has linear relations with test losses, while $L_2$ is correlated with test losses in a complicated nonlinear way. (3) LMN also reveals an intriguing phenomenon of the XOR network switching between two generalization solutions, while $L_2$ does not. Besides explaining grokking, we argue that LMN is a promising candidate as the neural network version of the Kolmogorov complexity since it explicitly considers local or conditioned linear computations aligned with the nature of modern artificial neural networks.