Author name cluster

Ziwei Ji

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

14 papers

2 author rows

NeurIPS Conference 2025 Conference Paper

Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation

Sangmin Bae
Yujin Kim
Reza Bayat
Sungnyun Kim
Jiyoun Ha
Tal Schuster
Adam Fisch
Hrayr Harutyunyan

Scaling language models unlocks impressive capabilities, but the accompanying computational and memory demands make both training and deployment expensive. Existing efficiency efforts typically target either parameter sharing or adaptive computation, leaving open the question of how to attain both simultaneously. We introduce Mixture-of-Recursions (MoR), a unified framework that combines the two axes of efficiency inside a single Recursive Transformer. MoR reuses a shared stack of layers across recursion steps to achieve parameter efficiency, while lightweight routers enable adaptive token-level thinking by dynamically assigning different recursion depths to individual tokens. This allows MoR to focus quadratic attention computation only among tokens still active at a given recursion depth, further improving memory access efficiency by selectively caching only their key-value pairs. Beyond these core mechanisms, we also propose a KV sharing variant that reuses KV pairs from the first recursion, specifically designed to further decrease memory footprint. Across model scales ranging from 135M to 1. 7B parameters, MoR forms a new Pareto frontier: at equal training FLOPs and smaller model sizes, it significantly lowers validation perplexity and improves few-shot accuracy, while delivering higher throughput compared with vanilla and existing recursive baselines.

PDF Details

ICLR Conference 2025 Conference Paper

Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA

Sangmin Bae
Adam Fisch
Hrayr Harutyunyan
Ziwei Ji
Seungyeon Kim 0001
Tal Schuster

Large language models (LLMs) are expensive to deploy. Parameter sharing offers a possible path towards reducing their size and cost, but its effectiveness in modern LLMs remains fairly limited. In this work, we revisit "layer tying" as form of parameter sharing in Transformers, and introduce novel methods for converting existing LLMs into smaller "Recursive Transformers" that share parameters across layers, with minimal loss of performance. Here, our Recursive Transformers are efficiently initialized from standard pretrained Transformers, but only use a single block of unique layers that is then repeated multiple times in a loop. We further improve performance by introducing Relaxed Recursive Transformers that add flexibility to the layer tying constraint via depth-wise low-rank adaptation (LoRA) modules, yet still preserve the compactness of the overall model. We show that our recursive models (e.g., recursive Gemma 1B) outperform both similar-sized vanilla pretrained models (such as TinyLlama 1.1B and Pythia 1B) and knowledge distillation baselines---and can even recover most of the performance of the original "full-size" model (e.g., Gemma 2B with no shared parameters). Finally, we propose Continuous Depth-wise Batching, a promising new inference paradigm enabled by the Recursive Transformer when paired with early exiting. In a theoretical analysis, we show that this has the potential to lead to significant (2-3$\times$) gains in inference throughput.

Details

NeurIPS Conference 2024 Conference Paper

ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models

Yuzhe Gu
Ziwei Ji
Wenwei Zhang
Chengqi Lyu
Dahua Lin
Kai Chen

Large language models (LLMs) exhibit hallucinations in long-form question-answering tasks across various domains and wide applications. Current hallucination detection and mitigation datasets are limited in domain and size, which struggle to scale due to prohibitive labor costs and insufficient reliability of existing hallucination annotators. To facilitate the scalable oversight of LLM hallucinations, this paper introduces an iterative self-training framework that simultaneously and progressively scales up the annotation dataset and improves the accuracy of the annotator. Based on the Expectation Maximization algorithm, in each iteration, the framework first applies an automatic hallucination annotation pipeline for a scaled dataset and then trains a more accurate annotator on the dataset. This new annotator is adopted in the annotation pipeline for the next iteration. Extensive experimental results demonstrate that the finally obtained hallucination annotator with only 7B parameters surpasses GPT-4 and obtains new state-of-the-art hallucination detection results on HaluEval and HalluQA by zero-shot inference. Such an annotator can not only evaluate the hallucination levels of various LLMs on the large-scale dataset but also help to mitigate the hallucination of LLMs generations, with the Natural Language Inference metric increasing from 25% to 37% on HaluEval.

PDF Details DOI

ICLR Conference 2024 Conference Paper

Think before you speak: Training Language Models With Pause Tokens

Sachin Goyal
Ziwei Ji
Ankit Singh Rawat
Aditya Krishna Menon
Sanjiv Kumar
Vaishnavh Nagarajan

Language models generate responses by producing a series of tokens in immediate succession: the $(K+1)^{\rm th}$ token is an outcome of manipulating $K$ hidden vectors per layer, one vector per preceding token. What if instead we were to let the model manipulate say, $K+10$ hidden vectors, before it outputs the $(K+1)^{\rm th}$ token? We operationalize this idea by performing training and inference on language models with a (learnable) $\textit{pause}$ token, a sequence of which is appended to the input prefix. We then delay extracting the model's outputs until the last pause token is seen, thereby allowing the model to process extra computation before committing to an answer. We empirically evaluate $\textit{pause-training}$ on decoder-only models of 1B and 130M parameters with causal pretraining on C4, and on downstream tasks covering reasoning, question-answering, general understanding and fact recall. Our main finding is that inference-time delays show gains when the model is both pre-trained and finetuned with delays. For the 1B model, we witness gains on 8 of 9 tasks, most prominently, a gain of $18\\%$ EM score on the QA task of SQuAD, $8\\%$ on CommonSenseQA and $1\\%$ accuracy on the reasoning task of GSM8k. Our work raises a range of conceptual and practical future research questions on making delayed next-token prediction a widely applicable new paradigm.

Details

ICLR Conference 2022 Conference Paper

Actor-critic is implicitly biased towards high entropy optimal policies

Yuzheng Hu
Ziwei Ji
Matus Telgarsky

We show that the simplest actor-critic method — a linear softmax policy updated with TD through interaction with a linear MDP, but featuring no explicit regularization or exploration — does not merely find an optimal policy, but moreover prefers high entropy optimal policies. To demonstrate the strength of this bias, the algorithm not only has no regularization, no projections, and no exploration like $\epsilon$-greedy, but is moreover trained on a single trajectory with no resets. The key consequence of the high entropy bias is that uniform mixing assumptions on the MDP, which exist in some form in all prior work, can be dropped: the implicit regularization of the high entropy bias is enough to ensure that all chains mix and an optimal policy is reached with high probability. As auxiliary contributions, this work decouples concerns between the actor and critic by writing the actor update as an explicit mirror descent, provides tools to uniformly bound mixing times within KL balls of policy space, and provides a projection-free TD analysis with its own implicit bias which can be run from an unmixed starting distribution.

Details

ICML Conference 2022 Conference Paper

Agnostic Learnability of Halfspaces via Logistic Loss

Ziwei Ji
Kwangjun Ahn
Pranjal Awasthi
Satyen Kale
Stefani Karp

We investigate approximation guarantees provided by logistic regression for the fundamental problem of agnostic learning of homogeneous halfspaces. Previously, for a certain broad class of “well-behaved” distributions on the examples, Diakonikolas et al. (2020) proved an tilde{Omega}(OPT) lower bound, while Frei et al. (2021) proved an tilde{O}(sqrt{OPT}) upper bound, where OPT denotes the best zero-one/misclassification risk of a homogeneous halfspace. In this paper, we close this gap by constructing a well-behaved distribution such that the global minimizer of the logistic risk over this distribution only achieves Omega(sqrt{OPT}) misclassification risk, matching the upper bound in (Frei et al. , 2021). On the other hand, we also show that if we impose a radial-Lipschitzness condition in addition to well-behaved-ness on the distribution, logistic regression on a ball of bounded radius reaches tilde{O}(OPT) misclassification risk. Our techniques also show for any well-behaved distribution, regardless of radial Lipschitzness, we can overcome the Omega(sqrt{OPT}) lower bound for logistic loss simply at the cost of one additional convex optimization step involving the hinge loss and attain tilde{O}(OPT) misclassification risk. This two-step convex optimization algorithm is simpler than previous methods obtaining this guarantee, all of which require solving O(log(1/OPT)) minimization problems.

Details

NeurIPS Conference 2022 Conference Paper

Reproducibility in Optimization: Theoretical Framework and Limits

Kwangjun Ahn
Prateek Jain
Ziwei Ji
Satyen Kale
Praneeth Netrapalli
Gil I Shamir

We initiate a formal study of reproducibility in optimization. We define a quantitative measure of reproducibility of optimization procedures in the face of noisy or error-prone operations such as inexact or stochastic gradient computations or inexact initialization. We then analyze several convex optimization settings of interest such as smooth, non-smooth, and strongly-convex objective functions and establish tight bounds on the limits of reproducibility in each setting. Our analysis reveals a fundamental trade-off between computation and reproducibility: more computation is necessary (and sufficient) for better reproducibility.

PDF Details

AAAI Conference 2021 Conference Paper

CrossNER: Evaluating Cross-Domain Named Entity Recognition

Zihan Liu
Yan Xu
Tiezheng Yu
Wenliang Dai
Ziwei Ji
Samuel Cahyawijaya
Andrea Madotto
Pascale Fung

Cross-domain named entity recognition (NER) models are able to cope with the scarcity issue of NER samples in target domains. However, most of the existing NER benchmarks lack domain-specialized entity types or do not focus on a certain domain, leading to a less effective cross-domain evaluation. To address these obstacles, we introduce a cross-domain NER dataset (CrossNER), a fully-labeled collection of NER data spanning over five diverse domains with specialized entity categories for different domains. Additionally, we also provide a domain-related corpus since using it to continue pre-training language models (domain-adaptive pre-training) is effective for the domain adaptation. We then conduct comprehensive experiments to explore the effectiveness of leveraging different levels of the domain corpus and pre-training strategies to do domain-adaptive pre-training for the crossdomain task. Results show that focusing on the fractional corpus containing domain-specialized entities and utilizing a more challenging pre-training strategy in domain-adaptive pre-training are beneficial for the NER domain adaptation, and our proposed method can consistently outperform existing cross-domain NER baselines. Nevertheless, experiments also illustrate the challenge of this cross-domain NER task. We hope that our dataset and baselines will catalyze research in the NER domain adaptation area. The code and data are available at https: //github. com/zliucr/CrossNER.

PDF Details

NeurIPS Conference 2021 Conference Paper

Early-stopped neural networks are consistent

Ziwei Ji
Justin Li
Matus Telgarsky

This work studies the behavior of shallow ReLU networks trained with the logistic loss via gradient descent on binary classification data where the underlying data distribution is general, and the (optimal) Bayes risk is not necessarily zero. In this setting, it is shown that gradient descent with early stopping achieves population risk arbitrarily close to optimal in terms of not just logistic and misclassification losses, but also in terms of calibration, meaning the sigmoid mapping of its outputs approximates the true underlying conditional distribution arbitrarily finely. Moreover, the necessary iteration, sample, and architectural complexities of this analysis all scale naturally with a certain complexity measure of the true conditional model. Lastly, while it is not shown that early stopping is necessary, it is shown that any classifier satisfying a basic local interpolation property is inconsistent.

PDF Details

ICML Conference 2021 Conference Paper

Fast margin maximization via dual acceleration

Ziwei Ji
Nathan Srebro
Matus Telgarsky

We present and analyze a momentum-based gradient method for training linear classifiers with an exponentially-tailed loss (e. g. , the exponential or logistic loss), which maximizes the classification margin on separable data at a rate of O(1/t^2). This contrasts with a rate of O(1/log(t)) for standard gradient descent, and O(1/t) for normalized gradient descent. The momentum-based method is derived via the convex dual of the maximum-margin problem, and specifically by applying Nesterov acceleration to this dual, which manages to result in a simple and intuitive method in the primal. This dual view can also be used to derive a stochastic variant, which performs adaptive non-uniform sampling via the dual variables.

Details

ICLR Conference 2021 Conference Paper

Generalization bounds via distillation

Daniel J. Hsu
Ziwei Ji
Matus Telgarsky
Lan Wang

This paper theoretically investigates the following empirical phenomenon: given a high-complexity network with poor generalization bounds, one can distill it into a network with nearly identical predictions but low complexity and vastly smaller generalization bounds. The main contribution is an analysis showing that the original network inherits this good generalization bound from its distillation, assuming the use of well-behaved data augmentation. This bound is presented both in an abstract and in a concrete form, the latter complemented by a reduction technique to handle modern computation graphs featuring convolutional layers, fully-connected layers, and skip connections, to name a few. To round out the story, a (looser) classical uniform convergence analysis of compression is also presented, as well as a variety of experiments on cifar and mnist demonstrating similar generalization performance between the original network and its distillation.

Details

NeurIPS Conference 2020 Conference Paper

Directional convergence and alignment in deep learning

Ziwei Ji
Matus Telgarsky

In this paper, we show that although the minimizers of cross-entropy and related classification losses are off at infinity, network weights learned by gradient flow converge in direction, with an immediate corollary that network predictions, training errors, and the margin distribution also converge. This proof holds for deep homogeneous networks — a broad class of networks allowing for ReLU, max-pooling, linear, and convolutional layers — and we additionally provide empirical support not just close to the theory (e. g. , the AlexNet), but also on non-homogeneous networks (e. g. , the DenseNet). If the network further has locally Lipschitz gradients, we show that these gradients also converge in direction, and asymptotically align with the gradient flow path, with consequences on margin maximization, convergence of saliency maps, and a few other settings. Our analysis complements and is distinct from the well-known neural tangent and mean-field theories, and in particular makes no requirements on network width and initialization, instead merely requiring perfect classification accuracy. The proof proceeds by developing a theory of unbounded nonsmooth Kurdyka-Łojasiewicz inequalities for functions definable in an o-minimal structure, and is also applicable outside deep learning.

PDF Details

ICLR Conference 2020 Conference Paper

Neural tangent kernels, transportation mappings, and universal approximation

Ziwei Ji
Matus Telgarsky
Ruicheng Xian

This paper establishes rates of universal approximation for the shallow neural tangent kernel (NTK): network weights are only allowed microscopic changes from random initialization, which entails that activations are mostly unchanged, and the network is nearly equivalent to its linearization. Concretely, the paper has two main contributions: a generic scheme to approximate functions with the NTK by sampling from transport mappings between the initial weights and their desired values, and the construction of transport mappings via Fourier transforms. Regarding the first contribution, the proof scheme provides another perspective on how the NTK regime arises from rescaling: redundancy in the weights due to resampling allows individual weights to be scaled down. Regarding the second contribution, the most notable transport mapping asserts that roughly $1 / \delta^{10d}$ nodes are sufficient to approximate continuous functions, where $\delta$ depends on the continuity properties of the target function. By contrast, nearly the same proof yields a bound of $1 / \delta^{2d}$ for shallow ReLU networks; this gap suggests a tantalizing direction for future work, separating shallow ReLU networks and their linearization.

Details

ICLR Conference 2020 Conference Paper

Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow ReLU networks

Ziwei Ji
Matus Telgarsky

Recent theoretical work has guaranteed that overparameterized networks trained by gradient descent achieve arbitrarily low training error, and sometimes even low test error. The required width, however, is always polynomial in at least one of the sample size $n$, the (inverse) target error $1/\epsilon$, and the (inverse) failure probability $1/\delta$. This work shows that $\widetilde{\Theta}(1/\epsilon)$ iterations of gradient descent with $\widetilde{\Omega}(1/\epsilon^2)$ training examples on two-layer ReLU networks of any width exceeding $\textrm{polylog}(n,1/\epsilon,1/\delta)$ suffice to achieve a test misclassification error of $\epsilon$. We also prove that stochastic gradient descent can achieve $\epsilon$ test error with polylogarithmic width and $\widetilde{\Theta}(1/\epsilon)$ samples. The analysis relies upon the separation margin of the limiting kernel, which is guaranteed positive, can distinguish between true labels and random labels, and can give a tight sample-complexity analysis in the infinite-width setting.

Details