Search

Kolmogorov-Arnold Networks (KAN) \cite{liu2024kan} were very recently proposed as a potential alternative to the prevalent architectural backbone of many deep learning models, the multi-layer perceptron (MLP). KANs have seen success in various tasks of AI for science, with their empirical efficiency and accuracy demostrated in function regression, PDE solving, and many more scientific problems. In this article, we revisit the comparison of KANs and MLPs, with emphasis on a theoretical perspective. On the one hand, we compare the representation and approximation capabilities of KANs and MLPs. We establish that MLPs can be represented using KANs of a comparable size. This shows that the approximation and representation capabilities of KANs are at least as good as MLPs. Conversely, we show that KANs can be represented using MLPs, but that in this representation the number of parameters increases by a factor of the KAN grid size. This suggests that KANs with a large grid size may be more efficient than MLPs at approximating certain functions. On the other hand, from the perspective of learning and optimization, we study the spectral bias of KANs compared with MLPs. We demonstrate that KANs are less biased toward low frequencies than MLPs. We highlight that the multi-level learning feature specific to KANs, i.e. grid extension of splines, improves the learning process for high-frequency components. Detailed comparisons with different choices of depth, width, and grid sizes of KANs are made, shedding some light on how to choose the hyperparameters in practice., Comment: 17 pages, 5 figures

Timely and adequate risk communication before natural hazards can reduce losses from extreme weather events and provide more resilient disaster preparedness. However, existing natural hazard risk communications have been abstract, ineffective, not immersive, and sometimes counterproductive. The implementation of virtual reality (VR) for natural hazard risk communication presents a promising alternative to the existing risk communication system by offering immersive and engaging experiences. However, it is still unknown how different modalities in VR could affect individuals' mitigation behaviors related to incoming natural hazards. In addition, it is also not clear how the repetitive risk communication of different modalities in the VR system leads to the effect of risk habituation. To fill the knowledge gap, we developed a VR system with a tornado risk communication scenario and conducted a mixed-design human subject experiment (N = 24). We comprehensively investigated our research using both quantitative and qualitative results., Comment: 14 pages, 10 figures, 3 tables, submitted to CHI'25

We present a deep HI survey at L-band conducted with the Five-hundred-meter Aperture Spherical radio Telescope (FAST) over the COSMOS field. This survey is strategically designed to overlap with the MIGHTEE COSMOS field, aiming to combine the sensitivity of the FAST and high-resolution of the MeerKAT. We observed the field with FAST for approximately 11 hours covering $\sim$2 square degrees, and reduced the raw data to HI spectral cubes over the frequency range 1310-1420 MHz. The FAST-HI data reach a median 3$\sigma$ column density of $N_{\rm HI}\sim2\times10^{17}$ cm$^{-2}$ over a 5 km s$^{-1}$ channel width, allowing for studies of the distribution of HI gas in various environments, such as in galaxies, the Circum-Galactic Medium (CGM) and Intergalactic Medium (IGM). We visually searched the spectral cubes for HI sources, and found a total of 80 HI detections, of which 56 have been cross-matched with the MIGHTEE-HI catalogue. With the cross-matched sources, we compare their HI masses and find that the total HI mass fraction in the IGM and CGM surrounding the galaxy pairs is statistically higher than the HI fraction surrounding the isolated galaxies by a difference of 13$\pm$4%, indicating that the CGM and IGM associated with interacting systems are richer in neutral hydrogen compared to those around isolated galaxies in the local Universe. We also describe several FAST-MeerKAT synergy projects, highlighting the full potential of exploiting both single-dish and interferometric observations to study the distribution and evolution of the diffuse HI gas., Comment: 13 pages, 14 figures; Accepted for publication in MNRAS; Minor corrections made at proof stage

Diffusion models are capable of generating photo-realistic images that combine elements which likely do not appear together in the training set, demonstrating the ability to \textit{compositionally generalize}. Nonetheless, the precise mechanism of compositionality and how it is acquired through training remains elusive. Inspired by cognitive neuroscientific approaches, we consider a highly reduced setting to examine whether and when diffusion models learn semantically meaningful and factorized representations of composable features. We performed extensive controlled experiments on conditional Denoising Diffusion Probabilistic Models (DDPMs) trained to generate various forms of 2D Gaussian bump images. We found that the models learn factorized but not fully continuous manifold representations for encoding continuous features of variation underlying the data. With such representations, models demonstrate superior feature compositionality but limited ability to interpolate over unseen values of a given feature. Our experimental results further demonstrate that diffusion models can attain compositionality with few compositional examples, suggesting a more efficient way to train DDPMs. Finally, we connect manifold formation in diffusion models to percolation theory in physics, offering insight into the sudden onset of factorized representation learning. Our thorough toy experiments thus contribute a deeper understanding of how diffusion models capture compositional structure in data., Comment: 11 pages, 6 figures, plus appendix, some content overlap with arXiv:2402.03305

This paper investigates a challenging problem of zero-shot learning in the multi-label scenario (MLZSL), wherein the model is trained to recognize multiple unseen classes within a sample (e.g., an image) based on seen classes and auxiliary knowledge, e.g., semantic information. Existing methods usually resort to analyzing the relationship of various seen classes residing in a sample from the dimension of spatial or semantic characteristics and transferring the learned model to unseen ones. However, they neglect the integrity of local and global features. Although the use of the attention structure will accurately locate local features, especially objects, it will significantly lose its integrity, and the relationship between classes will also be affected. Rough processing of global features will also directly affect comprehensiveness. This neglect will make the model lose its grasp of the main components of the image. Relying only on the local existence of seen classes during the inference stage introduces unavoidable bias. In this paper, we propose a novel and comprehensive visual-semantic framework for MLZSL, dubbed Epsilon, to fully make use of such properties and enable a more accurate and robust visual-semantic projection. In terms of spatial information, we achieve effective refinement by group aggregating image features into several semantic prompts. It can aggregate semantic information rather than class information, preserving the correlation between semantics. In terms of global semantics, we use global forward propagation to collect as much information as possible to ensure that semantics are not omitted. Experiments on large-scale MLZSL benchmark datasets NUS-Wide and Open-Images-v4 demonstrate that the proposed Epsilon outperforms other state-of-the-art methods with large margins., Comment: 11 pages, 6 figures. arXiv admin note: substantial text overlap with arXiv:2309.00923

A major challenge of AI + Science lies in their inherent incompatibility: today's AI is primarily based on connectionism, while science depends on symbolism. To bridge the two worlds, we propose a framework to seamlessly synergize Kolmogorov-Arnold Networks (KANs) and science. The framework highlights KANs' usage for three aspects of scientific discovery: identifying relevant features, revealing modular structures, and discovering symbolic formulas. The synergy is bidirectional: science to KAN (incorporating scientific knowledge into KANs), and KAN to science (extracting scientific insights from KANs). We highlight major new functionalities in the pykan package: (1) MultKAN: KANs with multiplication nodes. (2) kanpiler: a KAN compiler that compiles symbolic formulas into KANs. (3) tree converter: convert KANs (or any neural networks) to tree graphs. Based on these tools, we demonstrate KANs' capability to discover various types of physical laws, including conserved quantities, Lagrangians, symmetries, and constitutive laws., Comment: 27 pages, 14 figures

Training Transformer models on long sequences in a distributed setting poses significant challenges in terms of efficiency and scalability. Current methods are either constrained by the number of attention heads or excessive communication overheads. To address this problem, we propose WallFacer, a multi-dimensional distributed training system for long sequences, fostering an efficient communication paradigm and providing additional tuning flexibility for communication arrangements. Specifically, WallFacer introduces an extra parallel dimension to substantially reduce communication volume and avoid bandwidth bottlenecks. Through comprehensive experiments across diverse hardware environments and on both Natural Language Processing (NLP) and Computer Vision (CV) tasks, we demonstrate that our approach significantly surpasses state-of-the-art methods that support near-infinite sequence lengths, achieving performance improvements of up to 77.12% on GPT-style models and up to 114.33% on DiT (Diffusion Transformer) models.

Accurate flux density calibration is essential for precise analysis and interpretation of observations across different observation modes and instruments. In this research, we firstly introduce the flux calibration model incorporated in HIFAST pipeline, designed for processing HI 21-cm spectra. Furthermore, we investigate different calibration techniques and assess the dependence of the gain parameter on the time and environmental factors. A comparison is carried out in various observation modes (e.g. tracking and scanning modes) to determine the flux density gain ($G$), revealing insignificant discrepancies in $G$ among different methods. Long-term monitoring data shows a linear correlation between $G$ and atmospheric temperature. After subtracting the $G$--Temperature dependence, the dispersion of $G$ is reduced to $<$3% over a one-year time scale. The stability of the receiver response of FAST is considered sufficient to facilitate HI observations that can accommodate a moderate error in flux calibration (e.g., $>\sim5\%$) when utilizing a constant $G$ for calibration purposes. Our study will serve as a useful addition to the results provided by Jiang et al. (2020). Detailed measurement of $G$ for the 19 beams of FAST, covering the frequency range 1000 MHz -- 1500 MHz can be found on the HIFAST homepage: https://hifast.readthedocs.io/fluxgain., Comment: 14 pages, 15 figures, accepted by RAA

With the exceptional sensitivity of the Five-hundred-meter Aperture Spherical radio Telescope (FAST), we conducted observations of the neutral hydrogen (HI) in the circumgalactic medium of Andromeda's (M31) satellite galaxies, specifically Andromeda II, NGC 205, and NGC 185. Initially, three drift scans were executed for these satellites, with a detection limit of $4\times10^{18}$ cm$^{-2}$ ( approximately $1.88\times10^3 M_{\odot}$ of HI mass), followed by a more in-depth scan of a specific region. We discovered a C-shaped HI arc structure sharing a position and line-of-sight velocity similar to a stellar ring structure around Andromeda II, hinting at a potential connection with Andromeda II. In the context of NGC 205, we identified two mass concentrations in the northeast direction, which could be indicative of tidal streams resulting from the interaction between this galaxy and M31. These new lumps discovered could be very helpful in solving the missing interstellar medium (ISM) problem for NGC 205. Observations regarding NGC 185 are consistent with previous studies, and we did not detect any additional HI material around this galaxy. These observational results enhance our understanding of the evolution of these satellite galaxies and provide insight into their historical interactions with the galaxy M31., Comment: 9 pages, 7 figures, accepted by RAA

When a neural network can learn multiple distinct algorithms to solve a task, how does it "choose" between them during training? To approach this question, we take inspiration from ecology: when multiple species coexist, they eventually reach an equilibrium where some survive while others die out. Analogously, we suggest that a neural network at initialization contains many solutions (representations and algorithms), which compete with each other under pressure from resource constraints, with the "fittest" ultimately prevailing. To investigate this Survival of the Fittest hypothesis, we conduct a case study on neural networks performing modular addition, and find that these networks' multiple circular representations at different Fourier frequencies undergo such competitive dynamics, with only a few circles surviving at the end. We find that the frequencies with high initial signals and gradients, the "fittest," are more likely to survive. By increasing the embedding dimension, we also observe more surviving frequencies. Inspired by the Lotka-Volterra equations describing the dynamics between species, we find that the dynamics of the circles can be nicely characterized by a set of linear differential equations. Our results with modular addition show that it is possible to decompose complicated representations into simpler components, along with their basic interactions, to offer insight on the training dynamics of representations.

As the number of dementia patients rises, the need for accurate diagnostic procedures rises as well. Current methods, like using an MRI scan, rely on human input, which can be inaccurate. However, the decision logic behind machine learning algorithms and their outputs cannot be explained, as most operate in black-box models. Therefore, to increase the accuracy of diagnosing dementia through MRIs, a convolution neural network has been developed and trained using an open-source database of 6400 MRI scans divided into 4 dementia classes. The model, which attained a 98 percent validation accuracy, was shown to be well fit and able to generalize to new data. Furthermore, to aid in the visualization of the model output, an explainable AI algorithm was developed by visualizing the outputs of individual filters in each convolution layer, which highlighted regions of interest in the scan. These outputs do a great job of identifying the image features that contribute most to the model classification, thus allowing users to visualize and understand the results. Altogether, this combination of the convolution neural network and explainable AI algorithm creates a system that can be used in the medical field to not only aid in the proper classification of dementia but also allow everyone involved to visualize and understand the results., Comment: 4 pages, 4 figures

This article uses machine learning (ML) and explainable artificial intelligence (XAI) techniques to investigate the relationship between nutritional status and mortality rates associated with Alzheimers disease (AD). The Third National Health and Nutrition Examination Survey (NHANES III) database is employed for analysis. The random forest model is selected as the base model for XAI analysis, and the Shapley Additive Explanations (SHAP) method is used to assess feature importance. The results highlight significant nutritional factors such as serum vitamin B12 and glycated hemoglobin. The study demonstrates the effectiveness of random forests in predicting AD mortality compared to other diseases. This research provides insights into the impact of nutrition on AD and contributes to a deeper understanding of disease progression., Comment: 5 pages, 1 figure, 5 tables

How do transformers model physics? Do transformers model systems with interpretable analytical solutions, or do they create "alien physics" that are difficult for humans to decipher? We take a step in demystifying this larger puzzle by investigating the simple harmonic oscillator (SHO), $\ddot{x}+2\gamma \dot{x}+\omega_0^2x=0$, one of the most fundamental systems in physics. Our goal is to identify the methods transformers use to model the SHO, and to do so we hypothesize and evaluate possible methods by analyzing the encoding of these methods' intermediates. We develop four criteria for the use of a method within the simple testbed of linear regression, where our method is $y = wx$ and our intermediate is $w$: (1) Can the intermediate be predicted from hidden states? (2) Is the intermediate's encoding quality correlated with model performance? (3) Can the majority of variance in hidden states be explained by the intermediate? (4) Can we intervene on hidden states to produce predictable outcomes? Armed with these two correlational (1,2), weak causal (3) and strong causal (4) criteria, we determine that transformers use known numerical methods to model trajectories of the simple harmonic oscillator, specifically the matrix exponential method. Our analysis framework can conveniently extend to high-dimensional linear systems and nonlinear systems, which we hope will help reveal the "world model" hidden in transformers., Comment: 9 pages, 9 figures

Integrable partial differential equation (PDE) systems are of great interest in natural science, but are exceedingly rare and difficult to discover. To solve this, we introduce OptPDE, a first-of-its-kind machine learning approach that Optimizes PDEs' coefficients to maximize their number of conserved quantities, $n_{\rm CQ}$, and thus discover new integrable systems. We discover four families of integrable PDEs, one of which was previously known, and three of which have at least one conserved quantity but are new to the literature to the best of our knowledge. We investigate more deeply the properties of one of these novel PDE families, $u_t = (u_x+a^2u_{xxx})^3$. Our paper offers a promising schema of AI-human collaboration for integrable system discovery: machine learning generates interpretable hypotheses for possible integrable systems, which human scientists can verify and analyze, to truly close the discovery loop.

Inspired by the Kolmogorov-Arnold representation theorem, we propose Kolmogorov-Arnold Networks (KANs) as promising alternatives to Multi-Layer Perceptrons (MLPs). While MLPs have fixed activation functions on nodes ("neurons"), KANs have learnable activation functions on edges ("weights"). KANs have no linear weights at all -- every weight parameter is replaced by a univariate function parametrized as a spline. We show that this seemingly simple change makes KANs outperform MLPs in terms of accuracy and interpretability. For accuracy, much smaller KANs can achieve comparable or better accuracy than much larger MLPs in data fitting and PDE solving. Theoretically and empirically, KANs possess faster neural scaling laws than MLPs. For interpretability, KANs can be intuitively visualized and can easily interact with human users. Through two examples in mathematics and physics, KANs are shown to be useful collaborators helping scientists (re)discover mathematical and physical laws. In summary, KANs are promising alternatives for MLPs, opening opportunities for further improving today's deep learning models which rely heavily on MLPs., Comment: 48 pages, 20 figures. Codes are available at https://github.com/KindXiaoming/pykan

We present a statistical study of the properties of diffuse HI in ten nearby galaxies, comparing the HI detected by the single-dish telescope FAST (FEASTS program) and the interferometer VLA (THINGS program), respectively. The THINGS' observation missed HI with a median of 23% due to the short-spacing problem of interferometry and limited sensitivity. We extract the diffuse HI by subtracting the dense HI, which is obtained from the THINGS data with a uniform flux-density threshold, from the total HI detected by FAST. Among the sample, the median diffuse-HI fraction is 34%, and more diffuse HI is found in galaxies exhibiting more prominent tidal-interaction signatures. The diffuse HI we detected seems to be distributed in disk-like layers within a typical thickness of $1\,\text{kpc}$, different from the more halo-like diffuse HI detected around NGC 4631 in a previous study. Most of the diffuse HI is cospatial with the dense HI and has a typical column density of $10^{17.7}$-$10^{20.1}\,\text{cm}^{-2}$. The diffuse and dense HI exhibits a similar rotational motion, but the former lags by a median of 25% in at least the inner disks, and its velocity dispersions are typically twice as high. Based on a simplified estimation of circum-galactic medium properties and assuming pressure equilibrium, the volume density of diffuse HI appears to be constant within each individual galaxy, implying its role as a cooling interface. Comparing with existing models, these results are consistent with a possible link between tidal interactions, the formation of diffuse HI, and gas accretion., Comment: 45 pages, 23 figures. In press at ApJ. Data will be released at the FEASTS site upon publication

Scaling multi-dimensional transformers to long sequences is indispensable across various domains. However, the challenges of large memory requirements and slow speeds of such sequences necessitate sequence parallelism. All existing approaches fall under the category of embedded sequence parallelism, which are limited to shard along a single sequence dimension, thereby introducing significant communication overhead. However, the nature of multi-dimensional transformers involves independent calculations across multiple sequence dimensions. To this end, we propose Dynamic Sequence Parallelism (DSP) as a novel abstraction of sequence parallelism. DSP dynamically switches the parallel dimension among all sequences according to the computation stage with efficient resharding strategy. DSP offers significant reductions in communication costs, adaptability across modules, and ease of implementation with minimal constraints. Experimental evaluations demonstrate DSP's superiority over state-of-the-art embedded sequence parallelism methods by remarkable throughput improvements ranging from 32.2% to 10x, with less than 25% communication volume.

In recent times, the emergence of Large Language Models (LLMs) has resulted in increasingly larger model size, posing challenges for inference on low-resource devices. Prior approaches have explored offloading to facilitate low-memory inference but often suffer from efficiency due to I/O bottlenecks. To achieve low-latency LLMs inference on resource-constrained devices, we introduce HeteGen, a novel approach that presents a principled framework for heterogeneous parallel computing using CPUs and GPUs. Based on this framework, HeteGen further employs heterogeneous parallel computing and asynchronous overlap for LLMs to mitigate I/O bottlenecks. Our experiments demonstrate a substantial improvement in inference speed, surpassing state-of-the-art methods by over 317% at most., Comment: MLSys 2024

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., Comment: 12 pages, 6 figures

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., Comment: 10 pages, 8 figures, Published as a workshop paper at ICLR 2024

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., Comment: 24 pages

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., Comment: 13 pages, 9 figures

The Five-hundred-meter Aperture Spherical radio Telescope (FAST) has the largest aperture and a 19-beam L-band receiver, making it powerful for investigating the neutral hydrogen atomic gas (HI) in the universe. We present HiFAST (https://hifast.readthedocs.io), a dedicated, modular, and self-contained calibration and imaging pipeline for processing the HI data of FAST. The pipeline consists of frequency-dependent noise diode calibration, baseline fitting, standing wave removal using an FFT-based method, flux density calibration, stray radiation correction, and gridding to produce data cubes. These modules can be combined as needed to process the data from most FAST observation modes: tracking, drift scanning, On-The-Fly mapping, and most of their variants. With HiFAST, the RMS noises of the calibrated spectra from all 19 beams were only slightly (~ 5%) higher than the theoretical expectation. The results for the extended source M33 and the point sources are consistent with the results from Arecibo. The moment maps (0,1 and 2) of M33 agree well with the results from the Arecibo Galaxy Environment Survey (AGES) with a fractional difference of less than 10%. For a common sample of 221 sources with signal-to-noise ratio S/N >10 from the Arecibo Legacy Fast ALFA (ALFALFA) survey, the mean value of fractional difference in the integrated flux density, $S_{\mathrm{int}}$, between the two datasets is approximately 0.005 %, with a dispersion of 15.4%. Further checks on the integrated flux density of 23 sources with seven observations indicate that the variance in the flux density of the source with luminous objects ($S_\mathrm{int}$ $ > 2.5$ Jy km s$^{-1}$) is less than 5%. Our tests suggest that the FAST telescope, with the efficient, precise, and user-friendly pipeline HiFAST, will yield numerous significant scientific findings in the investigation of the HI in the universe., Comment: Accepted by SCPMA. 21 pages, 14 figures. The pipeline is accessible at https://hifast.readthedocs.io

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., Comment: ICLR 2024

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., Comment: 10 pages, 3 figures

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., Comment: 15 pages, 7 figures

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., Comment: 8 pages, 6 figures

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.

Neural scaling laws (NSL) refer to the phenomenon where model performance improves with scale. Sharma & Kaplan analyzed NSL using approximation theory and predict that MSE losses decay as $N^{-\alpha}$, $\alpha=4/d$, where $N$ is the number of model parameters, and $d$ is the intrinsic input dimension. Although their theory works well for some cases (e.g., ReLU networks), we surprisingly find that a simple 1D problem $y=x^2$ manifests a different scaling law ($\alpha=1$) from their predictions ($\alpha=4$). We opened the neural networks and found that the new scaling law originates from lottery ticket ensembling: a wider network on average has more "lottery tickets", which are ensembled to reduce the variance of outputs. We support the ensembling mechanism by mechanistically interpreting single neural networks, as well as studying them statistically. We attribute the $N^{-1}$ scaling law to the "central limit theorem" of lottery tickets. Finally, we discuss its potential implications for large language models and statistical physics-type theories of learning., Comment: 14 pages, 13 figures. Note from authors: the theory in this paper is questionable; we are trying our best to fix it. Empirical results still stand

This paper investigates a challenging problem of zero-shot learning in the multi-label scenario (MLZSL), wherein, the model is trained to recognize multiple unseen classes within a sample (e.g., an image) based on seen classes and auxiliary knowledge, e.g., semantic information. Existing methods usually resort to analyzing the relationship of various seen classes residing in a sample from the dimension of spatial or semantic characteristics, and transfer the learned model to unseen ones. But they ignore the effective integration of local and global features. That is, in the process of inferring unseen classes, global features represent the principal direction of the image in the feature space, while local features should maintain uniqueness within a certain range. This integrated neglect will make the model lose its grasp of the main components of the image. Relying only on the local existence of seen classes during the inference stage introduces unavoidable bias. In this paper, we propose a novel and effective group bi-enhancement framework for MLZSL, dubbed GBE-MLZSL, to fully make use of such properties and enable a more accurate and robust visual-semantic projection. Specifically, we split the feature maps into several feature groups, of which each feature group can be trained independently with the Local Information Distinguishing Module (LID) to ensure uniqueness. Meanwhile, a Global Enhancement Module (GEM) is designed to preserve the principal direction. Besides, a static graph structure is designed to construct the correlation of local features. Experiments on large-scale MLZSL benchmark datasets NUS-WIDE and Open-Images-v4 demonstrate that the proposed GBE-MLZSL outperforms other state-of-the-art methods with large margins., Comment: 11 pages, 8 figures

Large-scale language models have become increasingly challenging and expensive to train. Among various methods addressing this issue, Pipeline Parallelism has been widely employed to accommodate massive model weights within limited GPU memory. This paper introduces Hanayo, a wave-like pipeline parallelism strategy that boasts a concise structure and practical applicability, alongside a high-performance pipeline execution runtime to tackle the challenges of pipeline strategy implementation. Hanayo mitigates the issues of pipeline bubbles and excessive memory consumption prevalent in existing schemes, without resorting to model duplicates as in Chimera. Our evaluation, conducted on four distinct computing clusters and involving both GPT-like and BERT-like architectures with up to 32 GPUs, demonstrates up to a 30.4 \% increase in throughput compared to the state-of-the-art approach., Comment: 12 pages, 12 figures. This paper has been accepted by SC23(The International Conference for High Performance Computing, Networking, Storage and Analysis). The information can be seen at https://sc23.supercomputing.org/presentation/?id=pap352&sess=sess168

Do neural networks, trained on well-understood algorithmic tasks, reliably rediscover known algorithms for solving those tasks? Several recent studies, on tasks ranging from group arithmetic to in-context linear regression, have suggested that the answer is yes. Using modular addition as a prototypical problem, we show that algorithm discovery in neural networks is sometimes more complex. Small changes to model hyperparameters and initializations can induce the discovery of qualitatively different algorithms from a fixed training set, and even parallel implementations of multiple such algorithms. Some networks trained to perform modular addition implement a familiar Clock algorithm; others implement a previously undescribed, less intuitive, but comprehensible procedure which we term the Pizza algorithm, or a variety of even more complex procedures. Our results show that even simple learning problems can admit a surprising diversity of solutions, motivating the development of new tools for characterizing the behavior of neural networks across their algorithmic phase space., Comment: Accepted by NeurIPS 2023

Generative processes that involve solving differential equations, such as diffusion models, frequently necessitate balancing speed and quality. ODE-based samplers are fast but plateau in performance while SDE-based samplers deliver higher sample quality at the cost of increased sampling time. We attribute this difference to sampling errors: ODE-samplers involve smaller discretization errors while stochasticity in SDE contracts accumulated errors. Based on these findings, we propose a novel sampling algorithm called Restart in order to better balance discretization errors and contraction. The sampling method alternates between adding substantial noise in additional forward steps and strictly following a backward ODE. Empirically, Restart sampler surpasses previous SDE and ODE samplers in both speed and accuracy. Restart not only outperforms the previous best SDE results, but also accelerates the sampling speed by 10-fold / 2-fold on CIFAR-10 / ImageNet $64 \times 64$. In addition, it attains significantly better sample quality than ODE samplers within comparable sampling times. Moreover, Restart better balances text-image alignment/visual quality versus diversity than previous samplers in the large-scale text-to-image Stable Diffusion model pre-trained on LAION $512 \times 512$. Code is available at https://github.com/Newbeeer/diffusion_restart_sampling, Comment: Code is available at https://github.com/Newbeeer/diffusion_restart_sampling

Discovering conservation laws for a given dynamical system is important but challenging. In a theorist setup (differential equations and basis functions are both known), we propose the Sparse Invariant Detector (SID), an algorithm that auto-discovers conservation laws from differential equations. Its algorithmic simplicity allows robustness and interpretability of the discovered conserved quantities. We show that SID is able to rediscover known and even discover new conservation laws in a variety of systems. For two examples in fluid mechanics and atmospheric chemistry, SID discovers 14 and 3 conserved quantities, respectively, where only 12 and 2 were previously known to domain experts., Comment: The codes are available here: https://github.com/KindXiaoming/sid

We introduce Brain-Inspired Modular Training (BIMT), a method for making neural networks more modular and interpretable. Inspired by brains, BIMT embeds neurons in a geometric space and augments the loss function with a cost proportional to the length of each neuron connection. We demonstrate that BIMT discovers useful modular neural networks for many simple tasks, revealing compositional structures in symbolic formulas, interpretable decision boundaries and features for classification, and mathematical structure in algorithmic datasets. The ability to directly see modules with the naked eye can complement current mechanistic interpretability strategies such as probes, interventions or staring at all weights., Comment: Codes are available here: https://github.com/KindXiaoming/BIMT

Compositional Zero-shot Learning (CZSL) aims to recognize novel concepts composed of known knowledge without training samples. Standard CZSL either identifies visual primitives or enhances unseen composed entities, and as a result, entanglement between state and object primitives cannot be fully utilized. Admittedly, vision-language models (VLMs) could naturally cope with CZSL through tuning prompts, while uneven entanglement leads prompts to be dragged into local optimum. In this paper, we take a further step to introduce a novel Disentangled and Recurrent Prompt Tuning framework termed DRPT to better tap the potential of VLMs in CZSL. Specifically, the state and object primitives are deemed as learnable tokens of vocabulary embedded in prompts and tuned on seen compositions. Instead of jointly tuning state and object, we devise a disentangled and recurrent tuning strategy to suppress the traction force caused by entanglement and gradually optimize the token parameters, leading to a better prompt space. Notably, we develop a progressive fine-tuning procedure that allows for incremental updates to the prompts, optimizing the object first, then the state, and vice versa. Meanwhile, the optimization of state and object is independent, thus clearer features can be learned to further alleviate the issue of entangling misleading optimization. Moreover, we quantify and analyze the entanglement in CZSL and supplement entanglement rebalancing optimization schemes. DRPT surpasses representative state-of-the-art methods on extensive benchmark datasets, demonstrating superiority in both accuracy and efficiency.

Since diffusion models (DM) and the more recent Poisson flow generative models (PFGM) are inspired by physical processes, it is reasonable to ask: Can physical processes offer additional new generative models? We show that the answer is yes. We introduce a general family, Generative Models from Physical Processes (GenPhys), where we translate partial differential equations (PDEs) describing physical processes to generative models. We show that generative models can be constructed from s-generative PDEs (s for smooth). GenPhys subsume the two existing generative models (DM and PFGM) and even give rise to new families of generative models, e.g., "Yukawa Generative Models" inspired from weak interactions. On the other hand, some physical processes by default do not belong to the GenPhys family, e.g., the wave equation and the Schr\"{o}dinger equation, but could be made into the GenPhys family with some modifications. Our goal with GenPhys is to explore and expand the design space of generative models.

We propose the Quantization Model of neural scaling laws, explaining both the observed power law dropoff of loss with model and data size, and also the sudden emergence of new capabilities with scale. We derive this model from what we call the Quantization Hypothesis, where network knowledge and skills are "quantized" into discrete chunks ($\textbf{quanta}$). We show that when quanta are learned in order of decreasing use frequency, then a power law in use frequencies explains observed power law scaling of loss. We validate this prediction on toy datasets, then study how scaling curves decompose for large language models. Using language model gradients, we automatically decompose model behavior into a diverse set of skills (quanta). We tentatively find that the frequency at which these quanta are used in the training distribution roughly follows a power law corresponding with the empirical scaling exponent for language models, a prediction of our theory., Comment: 24 pages, 18 figures, NeurIPS 2023

We introduce a new family of physics-inspired generative models termed PFGM++ that unifies diffusion models and Poisson Flow Generative Models (PFGM). These models realize generative trajectories for $N$ dimensional data by embedding paths in $N{+}D$ dimensional space while still controlling the progression with a simple scalar norm of the $D$ additional variables. The new models reduce to PFGM when $D{=}1$ and to diffusion models when $D{\to}\infty$. The flexibility of choosing $D$ allows us to trade off robustness against rigidity as increasing $D$ results in more concentrated coupling between the data and the additional variable norms. We dispense with the biased large batch field targets used in PFGM and instead provide an unbiased perturbation-based objective similar to diffusion models. To explore different choices of $D$, we provide a direct alignment method for transferring well-tuned hyperparameters from diffusion models ($D{\to} \infty$) to any finite $D$ values. Our experiments show that models with finite $D$ can be superior to previous state-of-the-art diffusion models on CIFAR-10/FFHQ $64{\times}64$ datasets, with FID scores of $1.91/2.43$ when $D{=}2048/128$. In class-conditional setting, $D{=}2048$ yields current state-of-the-art FID of $1.74$ on CIFAR-10. In addition, we demonstrate that models with smaller $D$ exhibit improved robustness against modeling errors. Code is available at https://github.com/Newbeeer/pfgmpp, Comment: Code is available at https://github.com/Newbeeer/pfgmpp

Foundation models have impressive performance and generalization capabilities across a wide range of applications. The increasing size of the models introduces great challenges for the training. Tensor parallelism is a critical technique that is currently used in almost all foundation model training and has a significant impact on overall training performance. However, current tensor parallelism in machine learning frameworks misses optimization opportunities in fitting various interconnection topologies. In this work, we present ATP, an adaptive tensor parallelism framework for foundation models, which can automatically select the optimal parallel strategy on different interconnections. We propose column- and row-first tensor parallelism based on 2D device meshes and construct a search space. Combined with the hierarchical communication matrix, ATP can identify the optimal strategy in the search space. We also propose chunk-based overlapping to reduce communication overhead. Our evaluations show ATP consistently outperforms the state-of-the-art approaches for various model sizes and interconnects, achieving end-to-end training performance improvements of up to 37-64% on specific interconnects. Based on our theoretical model, the communication overhead of ATP decreases with scaling, indicating a qualitative leap forward.

We use the single-dish radio telescope FAST to map the HI in the tidally interacting NGC 4631 group with a resolution of 3.24$'$ (7 kpc), reaching a 5-$\sigma$ column density limit of $10^{17.9}$ cm$^{-2}$ assuming a line width of 20 km s$^{-1}$. Taking the existing interferometric HI image from the HALOGAS project of WSRT as reference, we are able to identify and characterize a significant excess of large-scale, low-density, and diffuse HI in the group. This diffuse HI extends for more than 120 kpc across, and accounts for more than one fourth of the total HI detected by FAST in and around the galaxy NGC 4631. In the region of the tidal tails, the diffuse HI has a typical column density above $10^{19.5}$ cm$^{-2}$, and is highly turbulent with a velocity dispersion around 50 km s$^{-1}$. It increases in column density with the dense HI, and tends to be associated with the kinematically ``hotter'' part of the dense HI. Through simple modeling, we find that the majority of the diffuse HI in the tail region is likely to induce cooling out of the hot IGM instead of evaporating or being radiatively ionized. Given these relations of gas in different phases, the diffuse HI may represent a condensing phase of the IGM. Active tidal interactions on-going and in the past may have produced the wide-spreading HI distribution, and triggered the gas accretion to NGC 4631 through the phase of the diffuse HI., Comment: 35 pages, 24 figures. Accepted for publication at ApJ. FEASTS site: http://kavli.pku.edu.cn/~jwang/FEASTS