Haozhe Shan

Haozhe Shan, Qianyi Li, Haim Sompolinsky · harvard

Continual learning (CL) enables animals to learn new tasks without erasing prior knowledge. CL in artificial neural networks (NNs) is challenging due to catastrophic forgetting, where new learning degrades performance on older tasks. While various techniques exist to mitigate forgetting, theoretical insights into when and why CL fails in NNs are lacking. Here, we present a statistical-mechanics theory of CL in deep, wide NNs, which characterizes the network's input-output mapping as it learns a sequence of tasks. It gives rise to order parameters (OPs) that capture how task relations and network architecture influence forgetting and anterograde interference, as verified by numerical evaluations. For networks with a shared readout for all tasks (single-head CL), the relevant-feature and rule similarity between tasks, respectively measured by two OPs, are sufficient to predict a wide range of CL behaviors. In addition, the theory predicts that increasing the network depth can effectively reduce interference between tasks, thereby lowering forgetting. For networks with task-specific readouts (multi-head CL), the theory identifies a phase transition where CL performance shifts dramatically as tasks become less similar, as measured by another task-similarity OP. While forgetting is relatively mild compared to single-head CL across all tasks, sufficiently low similarity leads to catastrophic anterograde interference, where the network retains old tasks perfectly but completely fails to generalize new learning. Our results delineate important factors affecting CL performance and suggest strategies for mitigating forgetting.

Haoyuan Shi, Xiancong Ren, Yingji Zhang et al.

Understanding how Vision-Language-Action (VLA) models transform multimodal knowledge into embodied control remains an open challenge. We present VLA-Trace, a progressive diagnostic framework that analyzes VLA models through a unified evidence chain from representation dynamics to causal control attribution and behavioral manifestation. It specifically combines cross-modal and checkpoint-drift centered kernel alignment (CKA) to trace representation evolution, attention knockout interventions to identify modality-specific control pathways, and rollout-level behavioral probes to examine grounding, shortcut dependence, and semantic following. Experiments on $π_{0.5}$ and OpenVLA reveal three key findings. First, the two models exhibit distinct modality-specific adaptation dynamics during VLA finetuning. Second, they rely on different multimodal routing strategies and layer-wise dependencies during action decoding. Third, although VLA policies excel at visually grounded trajectory generation, they remain limited in fine-grained semantic following. These findings highlight future directions for representation-preserving adaptation, causal VLA circuits, and compositional semantic control.

Haozhe Shan, Albert Gu, Zhong Meng et al.

Online speech recognition, where the model only accesses context to the left, is an important and challenging use case for ASR systems. In this work, we investigate augmenting neural encoders for online ASR by incorporating structured state-space sequence models (S4), a family of models that provide a parameter-efficient way of accessing arbitrarily long left context. We performed systematic ablation studies to compare variants of S4 models and propose two novel approaches that combine them with convolutions. We found that the most effective design is to stack a small S4 using real-valued recurrent weights with a local convolution, allowing them to work complementarily. Our best model achieves WERs of 4.01%/8.53% on test sets from Librispeech, outperforming Conformers with extensively tuned convolution.

Haozhe Shan, Xiancong Ren, Han Dong et al.

While large vision-language models (VLMs) are increasingly adopted as the perceptual backbone for embodied agents, existing benchmarks often rely on question-answering or multiple-choice formats. These protocols allow models to exploit linguistic priors rather than demonstrating genuine visual grounding. To address this, we present EPIC-Bench, Embodied PerceptIon BenChmark, a fine-grained grounding benchmark designed to systematically evaluate the visual perceptual capabilities of VLMs in real-world embodied environments. Comprising 6.6k meticulously annotated tuples (Image, Text, Mask), EPIC-Bench spans 23 fine-grained tasks across three core stages of the embodied interaction pipeline: Target Localization, Navigation, and Manipulation. Extensive evaluations of over 89 leading VLMs reveal that while advanced reasoning models show promise, current VLMs universally struggle with complex visual-text alignment for physical interactions. Specifically, models exhibit critical bottlenecks in multi-target counting, part-whole relationship understanding, and affordance region detection. EPIC-Bench provides a robust foundation and actionable insights for advancing the next generation of vision-driven embodied models.

Yi Zhang, Che Liu, Xiancong Ren et al.

Developing a universal and versatile embodied intelligence system presents two primary challenges: the critical embodied data bottleneck, where real-world data is scarce and expensive, and the algorithmic inefficiency of existing methods, which are resource-prohibitive. To address these limitations, we introduce Deliberate Practice Policy Optimization (DPPO), a metacognitive ``Metaloop'' training framework that dynamically alternates between supervised fine-tuning (competence expansion) and reinforcement learning (skill refinement). This enables automatic weakness identification and targeted resource allocation, specifically designed to maximize learning efficiency from sparse, finite data. Theoretically, DPPO can be formalised as a unified preference-learning framework. Empirically, training a vision-language embodied model with DPPO, referred to as Pelican-VL 1.0, yields a 20.3% performance improvement over the base model and surpasses open-source models at the 100B-parameter scale by 10.6%. We are open-sourcing both the models and code, providing the first systematic framework that alleviates the data and resource bottleneck and enables the community to build versatile embodied agents efficiently.

Haozhe Shan, Sun Minni, Lea Duncker

The ability to continually learn, retain and deploy skills to accomplish goals is a key feature of intelligent and efficient behavior. However, the neural mechanisms facilitating the continual learning and flexible (re-)composition of skills remain elusive. Here, we study continual learning and the compositional reuse of learned computations in recurrent neural network (RNN) models using a novel two-system approach: one system that infers what computation to perform, and one that implements how to perform it. We focus on a set of compositional cognitive tasks commonly studied in neuroscience. To construct the what system, we first show that a large family of tasks can be systematically described by a probabilistic generative model, where compositionality stems from a shared underlying vocabulary of discrete task epochs. The shared epoch structure makes these tasks inherently compositional. We first show that this compositionality can be systematically described by a probabilistic generative model. Furthermore, We develop an unsupervised online learning approach that can learn this model on a single-trial basis, building its vocabulary incrementally as it is exposed to new tasks, and inferring the latent epoch structure as a time-varying computational context within a trial. We implement the how system as an RNN whose low-rank components are composed according to the context inferred by the what system. Contextual inference facilitates the creation, learning, and reuse of low-rank RNN components as new tasks are introduced sequentially, enabling continual learning without catastrophic forgetting. Using an example task set, we demonstrate the efficacy and competitive performance of this two-system learning framework, its potential for forward and backward transfer, as well as fast compositional generalization to unseen tasks.

Haozhe Shan, Blake Bordelon

The training dynamics and generalization properties of neural networks (NN) can be precisely characterized in function space via the neural tangent kernel (NTK). Structural changes to the NTK during training reflect feature learning and underlie the superior performance of networks outside of the static kernel regime. In this work, we seek to theoretically understand kernel alignment, a prominent and ubiquitous structural change that aligns the NTK with the target function. We first study a toy model of kernel evolution in which the NTK evolves to accelerate training and show that alignment naturally emerges from this demand. We then study alignment mechanism in deep linear networks and two layer ReLU networks. These theories provide good qualitative descriptions of kernel alignment and specialization in practical networks and identify factors in network architecture and data structure that drive kernel alignment. In nonlinear networks with multiple outputs, we identify the phenomenon of kernel specialization, where the kernel function for each output head preferentially aligns to its own target function. Together, our results provide a mechanistic explanation of how kernel alignment emerges during NN training and a normative explanation of how it benefits training.

Haozhe Shan, Peggy Mason

Behaviors of several laboratory animals can be modeled as sequences of stereotyped behaviors, or behavioral motifs. However, identifying such motifs is a challenging problem. Behaviors have a multi-scale structure: the animal can be simultaneously performing a small-scale motif and a large-scale one (e.g. \textit{chewing} and \textit{feeding}). Motifs are compositional: a large-scale motif is a chain of smaller-scale ones, folded in (some behavioral) space in a specific manner. We demonstrate an approach which captures these structures, using rat locomotor data as an example. From the same dataset, we used a preprocessing procedure to create different versions, each describing motifs of a different scale. We then trained several Hidden Markov Models (HMMs) in parallel, one for each dataset version. This approach essentially forced each HMM to learn motifs on a different scale, allowing us to capture behavioral structures lost in previous approaches. By comparing HMMs with models representing different null hypotheses, we found that rat locomotion was composed of distinct motifs from second scale to minute scale. We found that transitions between motifs were modulated by rats' location in the environment, leading to non-Markovian transitions. To test the ethological relevance of motifs we discovered, we compared their usage between rats with differences in a high-level trait, prosociality. We found that these rats had distinct motif repertoires, suggesting that motif usage statistics can be used to infer internal states of rats. Our method is therefore an efficient way to discover multi-scale, compositional structures in animal behaviors. It may also be applied as a sensitive assay for internal states.