Research on learning, adaptation, and generalization.
2026
Advances in Cognitive Systems 2026 · 2026 · 🏆 Ada Lovelace Award for Excellent Paper
Zekun Wang, Karthik Singaravadivelan, Christopher J. MacLellan
A diffusion model's noise levels reveal a hierarchy of concepts, connecting image generation with how people organize knowledge.
ICML 2026 Mechanistic Interpretability Workshop · 2026 · Poster
Ruixuan Deng, Zehao Jin, Zekun Wang, Zihan Dong
Separating a compact, dense scaffold from sparse features helps explain what sparse autoencoders capture—and what they represent poorly.
arXiv 2026 · 2026
Zekun Wang, Anant Gupta, Tianyi Zhu, Christopher J. MacLellan
Recovers query-specific concepts from a pretrained diffusion model's time-indexed scores via gradient ascent, then composes them as a product-of-experts teacher for test-time generalization.
arXiv 2026 · 2026
Zekun Wang, Anant Gupta, Zihan Dong, Christopher J. MacLellan
SCoL post-trains an LLM via meta-RL to emit textual instructions selecting which Transformer layers to LoRA-update from streaming context, balancing acquisition against forgetting.
AAAI 2026 · 2026
Zekun Wang, Sashank Varma
ResNet-50's accuracy on Dehaene's odd-one-out tasks for geometry and number, tracked across ImageNet training checkpoints, mirrors developmental trajectories observed in children.
Findings of ACL 2026 · 2026
Aryan Roy, Zekun Wang, Christopher J. MacLellan
Compares matched Llama text and Llama Vision models on text-only prompts to test whether multimodal pretraining yields more human-like sensitivity to word concreteness.
NeurIPS 2026 · 2026
Zekun Wang, Anant Gupta, Christopher J. MacLellan
Combines generative replay with a Fisher-metric trust region for continual diffusion training, and shows the update reduces to a one-step MAML form.
Findings of ACL 2026 · 2026
Karthik Singaravadivelan, Anant Gupta, Zekun Wang, Christopher J. MacLellan
CobwebTM adapts the Cobweb concept-formation algorithm to dense document embeddings, building a topic hierarchy incrementally as documents stream in.
ICLR 2026 · 2026
Zekun Wang, Anant Gupta, Zihan Dong, Christopher J. MacLellan
Proposes a rank-1 Fisher approximation for EWC that exploits the collinear per-sample gradients of diffusion models at low SNR, paired with generative replay for continual image generation.
CogSci 2026 · 2026
Christopher J. MacLellan, Karthik Singaravadivelan, Xin Lian, Zekun Wang, Pat Langley
Extends Cobweb with pointwise mutual information to model human categorization, fitting classic Hayes-Roth, Medin & Schaffer, and Smith & Minda data better than Anderson's rational model.
2025
Cognitive Systems Research 92 (2025), 101371 · 2025
Xin Lian, Zekun Wang, Christopher J. MacLellan
Using concept formation to study efficient and scalable masked word prediction.
NAACL 2025 · 2025
Ziqiao Ma, Zekun Wang, Joyce Chai
Trains GPT-2 from scratch with a teacher model giving demonstrations on student trials, scored by an age-conditioned reward, to accelerate word acquisition.
CogSci 2025 · 2025
Zekun Wang, Sashank Varma
Tests CNNs, vision transformers, and vision-language models on a 43-concept odd-one-out task and finds transformers match children's geometric sensitivity while VLMs underperform.
Advances in Cognitive Systems 12 (2025) · 2025
Anant Gupta, Karthik Singaravadivelan, Zekun Wang
Organizes sentence embeddings into a Cobweb prototype tree and retrieves documents via coarse-to-fine traversal, matching dot-product retrieval and staying robust when GPT-2 embeddings degrade.
NeuS 2025 · 2025
Zekun Wang, Ethan L. Haarer, Nicki Barari, Christopher J. MacLellan
Introduces taxonomic networks as a shared representation that pairs a symbolic learner and a neural learner, allowing models to be translated between paradigms.
NeurIPS 2025 · 2025
Zekun Wang, Ethan Haarer, Tianyi Zhu, Zhiyi Dai, Christopher J. MacLellan
Trains a VAE with a complete binary-tree mixture-of-Gaussians prior to discover hierarchical prototype clusters from unlabeled images without specifying class counts.
2024
LREC-COLING 2024 · 2024
Oana Ignat, Zhijing Jin, Artem Abzaliev, Laura Biester, Santiago Castro, Naihao Deng, Xinyi Gao, Aylin Gunal, Jacky He, Ashkan Kazemi, Muhammad Khalifa, Namho Koh, Andrew Lee, Siyang Liu, Do June Min, Shinka Mori, Joan Nwatu, Veronica Perez-Rosas, Siqi Shen, Zekun Wang, Winston Wu, Rada Mihalcea
This position paper catalogs 14 NLP research areas and 45 specific directions that remain open and tractable for academic labs despite LLM progress.