Searcharxiv⌕ Search

arXiv subjects

Qihan Wang

Publications and source records attributed to Qihan Wang.

10 recordsLinked to original sources
arXiv2026

Simulating Human Memory with Language Models

Language models are increasingly being deployed as user simulators, but their memory is far more reliable than that of real users. To measure this gap, we run a series of classic memory experiments from psychology on both humans and language models. Across tasks, we find that out-of-the-box language models exhibit better memory than humans, even when prompted to imitate human behavior. We then show that better prompting strategies and the use of a compactor can cause language models to forget content in a more human-like way. Using these methods, we show preliminary evidence that language models with human-like memory constraints can function as more effective user simulators in a downstream education task. Finally, we release human reference data and benchmarks to support future work on simulating human memory with language models.

cs.CL↗
arXiv2025

Multilingual Prompting for Improving LLM Generation Diversity

Large Language Models (LLMs) are known to lack cultural representation and overall diversity in their generations, from expressing opinions to answering factual questions. To mitigate this problem, we propose multilingual prompting: a prompting method which generates several variations of a base prompt with added cultural and linguistic cues from several cultures, generates responses, and then combines the results. Building on evidence that LLMs have language-specific knowledge, multilingual prompting seeks to increase diversity by activating a broader range of cultural knowledge embedded in model training data. Through experiments across multiple models (GPT-4o, GPT-4o-mini, LLaMA 70B, and LLaMA 8B), we show that multilingual prompting consistently outperforms existing diversity-enhancing techniques such as high-temperature sampling, step-by-step recall, and persona prompting. Further analyses show that the benefits of multilingual prompting vary between high and low resource languages and across model sizes, and that aligning the prompting language with cultural cues reduces hallucination about culturally-specific information.

cs.CL↗
arXiv2025

Lyapunov Exponent and Stochastic Stability for Infinitely Renormalizable Lorenz Maps

We prove that infinitely renormalizable contracting Lorenz maps with bounded geometry or the so-called {\it a priori bounds} satisfies the slow recurrence condition to the singular point $c$ at its two critical values $c_1^-$ and $c_1^+$. As the first application, we show that the pointwise Lyapunov exponent at $c_1^-$ and $c_1^+$ equals 0. As the second application, we show that such maps are stochastically stable.

math.DS↗
arXiv2023

Exploring Musical, Lyrical, and Network Dimensions of Music Sharing Among Depression Individuals

Depression has emerged as a significant mental health concern due to a variety of factors, reflecting broader societal and individual challenges. Within the digital era, social media has become an important platform for individuals navigating through depression, enabling them to express their emotional and mental states through various mediums, notably music. Specifically, their music preferences, manifested through sharing practices, inadvertently offer a glimpse into their psychological and emotional landscapes. This work seeks to study the differences in music preferences between individuals diagnosed with depression and non-diagnosed individuals, exploring numerous facets of music, including musical features, lyrics, and musical networks. The music preferences of individuals with depression through music sharing on social media, reveal notable differences in musical features and topics and language use of lyrics compared to non-depressed individuals. We find the network information enhances understanding of the link between music listening patterns. The result highlights a potential echo-chamber effect, where depression individual's musical choices may inadvertently perpetuate depressive moods and emotions. In sum, this study underscores the significance of examining music's various aspects to grasp its relationship with mental health, offering insights for personalized music interventions and recommendation algorithms that could benefit individuals with depression.

cs.CY↗
arXiv2023

How much pre-training is enough to discover a good subnetwork?

Neural network pruning is useful for discovering efficient, high-performing subnetworks within pre-trained, dense network architectures. More often than not, it involves a three-step process -- pre-training, pruning, and re-training -- that is computationally expensive, as the dense model must be fully pre-trained. While previous work has revealed through experiments the relationship between the amount of pre-training and the performance of the pruned network, a theoretical characterization of such dependency is still missing. Aiming to mathematically analyze the amount of dense network pre-training needed for a pruned network to perform well, we discover a simple theoretical bound in the number of gradient descent pre-training iterations on a two-layer, fully-connected network, beyond which pruning via greedy forward selection [61] yields a subnetwork that achieves good training error. Interestingly, this threshold is shown to be logarithmically dependent upon the size of the dataset, meaning that experiments with larger datasets require more pre-training for subnetworks obtained via pruning to perform well. Lastly, we empirically validate our theoretical results on a multi-layer perceptron trained on MNIST.

stat.ML↗
arXiv2022

The Magic of Slow-to-Fast and Constant: Evaluating Time Perception of Progress Bars by Bayesian Model

Objective: We aimed to use adaptive psychophysics methods, which is a Bayesian Model, to measure users' time perception of various progress bar quantitatively. Background: Progress bar informs users about the status of ongoing processes. Progress bars frequently display nonuniform speed patterns, such as acceleration and deceleration. However, which progress bar is perceived faster remain unclear. Methods: We measured the point of subject equality (PSE) of the constant progress bar toward four different 5-second progress bars with a non-constant speed. To measure PSE, in each trial, a constant progress bar and a non-constant progress bar were presented to participants. Participants needed to judge which one is shorter. Based on their choice, the model generated the time duration of constant progress bar in next trial. After 40 trials for each non-constant progress bar, the PSE was calculated by the model. Eye tracking was recorded during the experiment.Results: Our results show that the constant progress bar and speed-up progress bar are perceived to be faster. The anchoring effect fits the results of our study, indicating that the final part of the progress bar is more important for time perception. Moreover, the eye-tracking results indicate that the progress bar is perceived to be slower is related to the overload of cognitive resources.Conclusion: The constant progress bar and speed-up progress bar are perceived as the quickest. Application: The results suggest that UX design can use constant or speed-up progress bar, in order to improve user experience in waiting.

cs.HC↗
arXiv2022

LOFT: Finding Lottery Tickets through Filter-wise Training

Recent work on the Lottery Ticket Hypothesis (LTH) shows that there exist ``\textit{winning tickets}'' in large neural networks. These tickets represent ``sparse'' versions of the full model that can be trained independently to achieve comparable accuracy with respect to the full model. However, finding the winning tickets requires one to \emph{pretrain} the large model for at least a number of epochs, which can be a burdensome task, especially when the original neural network gets larger. In this paper, we explore how one can efficiently identify the emergence of such winning tickets, and use this observation to design efficient pretraining algorithms. For clarity of exposition, our focus is on convolutional neural networks (CNNs). To identify good filters, we propose a novel filter distance metric that well-represents the model convergence. As our theory dictates, our filter analysis behaves consistently with recent findings of neural network learning dynamics. Motivated by these observations, we present the \emph{LOttery ticket through Filter-wise Training} algorithm, dubbed as \textsc{LoFT}. \textsc{LoFT} is a model-parallel pretraining algorithm that partitions convolutional layers by filters to train them independently in a distributed setting, resulting in reduced memory and communication costs during pretraining. Experiments show that \textsc{LoFT} $i)$ preserves and finds good lottery tickets, while $ii)$ it achieves non-trivial computation and communication savings, and maintains comparable or even better accuracy than other pretraining methods.

cs.LG↗
arXiv2021

Mitigating deep double descent by concatenating inputs

The double descent curve is one of the most intriguing properties of deep neural networks. It contrasts the classical bias-variance curve with the behavior of modern neural networks, occurring where the number of samples nears the number of parameters. In this work, we explore the connection between the double descent phenomena and the number of samples in the deep neural network setting. In particular, we propose a construction which augments the existing dataset by artificially increasing the number of samples. This construction empirically mitigates the double descent curve in this setting. We reproduce existing work on deep double descent, and observe a smooth descent into the overparameterized region for our construction. This occurs both with respect to the model size, and with respect to the number epochs.

cs.LG↗