arXiv subjects

Trung Dang

Publications and source records attributed to Trung Dang.

At least 19 recordsLinked to original sources

arXiv2026

Pricing Pandora's Boxes: Revenue Maximization in Sequential Information Acquisition

We study a mechanism design problem in which a seller controls access to information about a set of stochastic alternatives, and a buyer sequentially acquires information in order to choose a single alternative with high value. The value distributions of the alternatives are known to both parties. The seller posts non adaptive prices for revealing each alternative's realized value, and the buyer responds optimally by following a Pandora's Box strategy: deciding which alternatives to inspect and when to stop by accepting the best inspected alternative. The seller's goal is to maximize his expected revenue, i.e. the total payment collected from all inspections. We study the revenue objective through the lens of simplicity versus optimality. Our main result is that a simple and efficiently computable pricing scheme obtains a 4 approximation in the worst case to the optimal revenue. This pricing rule equalizes the Weitzman indices across all alternatives. In contrast, we show that equalizing the prices themselves can be an unbounded factor worse than the optimum. Furthermore, for several natural special cases, including identically distributed alternatives and monotone hazard rate distributions, we fully characterize the optimal pricing. Finally, we also study a variant of our model under optional inspection, where the buyer may select an alternative without observing its realization. In this setting, we obtain an n/(n-1) approximation for the special case of n identically distributed alternatives, as well as a 2 approximation for the special case where each alternative's value distribution has support size two. Overall, our results highlight both the computational challenges and the power of simple pricing schemes in selling information to a sequential searcher.

cs.DS↗

arXiv2026

Fair Division Under Inaccurate Preferences

The fair allocation of scarce resources is a central problem in mathematics, computer science, operations research, and economics. While much of the fair-division literature assumes that individuals have underlying cardinal preferences, eliciting exact numerical values is often cognitively burdensome and prone to inaccuracies. A growing body of work in fair division addresses this challenge by assuming access only to ordinal preferences. However, the restricted expressiveness of ordinal preferences makes it challenging to quantify and optimize cardinal fairness objectives such as envy. In this paper, we explore the broad landscape of fair division of indivisible items given inaccurate cardinal preferences, with a focus on minimizing envy. We consider various settings based on whether the true preferences of the agents are stochastic or worst-case, and whether the inaccuracies, modeled as additive noise, are stochastic or worst-case. When the true preferences are stochastic, we show that envy-free allocations can be computed with high probability; this is achieved both in the setting with stochastic and worst-case noise. This generalizes a notable result in stochastic fair division, which establishes a similar guarantee, albeit in the absence of any noise. When the true preferences are worst-case, and the noise is bounded, we analyze the maximum envy achieved by the Round-Robin algorithm. This bound is shown to be tight for deterministic algorithms, and applications of this bound are provided. Lastly, we consider a setting with worst-case preferences and noise, where the true preferences for each item are revealed upon its allocation. Here, we give an efficient online algorithm that guarantees logarithmic maximum envy with high probability. This result generalizes a known result from algorithmic discrepancy to a setting with noisy input data.

cs.GT↗

arXiv2026

TADA: A Generative Framework for Speech Modeling via Text-Acoustic Dual Alignment

Modern Text-to-Speech (TTS) systems increasingly leverage Large Language Model (LLM) architectures to achieve scalable, high-fidelity, zero-shot generation. However, these systems typically rely on fixed-frame-rate acoustic tokenization, resulting in speech sequences that are significantly longer than, and asynchronous with their corresponding text. Beyond computational inefficiency, this sequence length disparity often triggers hallucinations in TTS and amplifies the modality gap in spoken language modeling (SLM). In this paper, we propose a novel tokenization scheme that establishes one-to-one synchronization between continuous acoustic features and text tokens, enabling unified, single-stream modeling within an LLM. We demonstrate that these synchronous tokens maintain high-fidelity audio reconstruction and can be effectively modeled in a latent space by a large language model with a flow matching head. Moreover, the ability to seamlessly toggle speech modality within the context enables text-only guidance--a technique that blends logits from text-only and text-speech modes to flexibly bridge the gap toward text-only LLM intelligence. Experimental results indicate that our approach achieves performance competitive with state-of-the-art TTS and SLM systems while virtually eliminating content hallucinations and preserving linguistic integrity, all at a significantly reduced inference cost.

cs.SD↗

arXiv2025

Commitment Gap via Correlation Gap

Selection problems with costly information, dating back to Weitzman's Pandora's Box problem, have received much attention recently. We study the general model of Costly Information Combinatorial Selection (CICS) that was recently introduced by Chawla et al. [2024] and Bowers et al. [2025]. In this problem, a decision maker needs to select a feasible subset of stochastic variables, and can only learn information about their values through a series of costly steps, modeled by a Markov decision process. The algorithmic objective is to maximize the total value of the selection minus the cost of information acquisition. However, determining the optimal algorithm is known to be a computationally challenging problem. To address this challenge, previous approaches have turned to approximation algorithms by considering a restricted class of committing policies that simplify the decision-making aspects of the problem and allow for efficient optimization. This motivates the question of bounding the commitment gap, measuring the worst case ratio in the performance of the optimal committing policy and the overall optimal. In this work, we obtain improved bounds on the commitment gap of CICS through a reduction to a simpler problem of Bayesian Combinatorial Selection where information is free. By establishing a close relationship between these problems, we are able to relate the commitment gap of CICS to ex ante free-order prophet inequalities. As a consequence, we obtain improved efficient approximations for arbitrary instances of the CICS under various feasibility constraints.

cs.DS↗

arXiv2025

Multi-Unit Combinatorial Prophet Inequalities

We consider a combinatorial auction setting where buyers have fractionally subadditive (XOS) valuations over the items and the seller's objective is to maximize the social welfare. A prophet inequality in this setting bounds the competitive ratio of sequential allocation (often using item pricing) against the hindsight optimum. We study the dependence of the competitive ratio on the number of copies, $k$, of each item. We show that the multi-unit combinatorial setting is strictly harder than its single-item counterpart in that there is a gap between the competitive ratios achieved by static item pricings in the two settings. However, if the seller is allowed to change item prices dynamically, it becomes possible to asymptotically match the competitive ratio of a single-item static pricing. We also develop a new non-adaptive anonymous multi-unit combinatorial prophet inequality where the item prices are determined up front but increase as the item supply decreases. Setting the item prices in our prophet inequality requires minimal information about the buyers' value distributions -- merely (an estimate of) the expected social welfare accrued by each item in the hindsight optimal solution suffices. Our non-adaptive pricing achieves a competitive ratio that increases strictly as a function of the item supply $k$.

cs.GT↗

arXiv2024

Zero-Shot Text-to-Speech from Continuous Text Streams

Existing zero-shot text-to-speech (TTS) systems are typically designed to process complete sentences and are constrained by the maximum duration for which they have been trained. However, in many streaming applications, texts arrive continuously in short chunks, necessitating instant responses from the system. We identify the essential capabilities required for chunk-level streaming and introduce LiveSpeech 2, a stream-aware model that supports infinitely long speech generation, text-audio stream synchronization, and seamless transitions between short speech chunks. To achieve these, we propose (1) adopting Mamba, a class of sequence modeling distinguished by linear-time decoding, which is augmented by cross-attention mechanisms for conditioning, (2) utilizing rotary positional embeddings in the computation of cross-attention, enabling the model to process an infinite text stream by sliding a window, and (3) decoding with semantic guidance, a technique that aligns speech with the transcript during inference with minimal overhead. Experimental results demonstrate that our models are competitive with state-of-the-art language model-based zero-shot TTS models, while also providing flexibility to support a wide range of streaming scenarios.

cs.SD↗

arXiv2024

Robust Max Selection

We introduce a new model to study algorithm design under unreliable information, and apply this model for the problem of finding the uncorrupted maximum element of a list containing $n$ elements, among which are $k$ corrupted elements. Under our model, algorithms can perform black-box comparison queries between any pair of elements. However, queries regarding corrupted elements may have arbitrary output. In particular, corrupted elements do not need to behave as any consistent values, and may introduce cycles in the elements' ordering. This imposes new challenges for designing correct algorithms under this setting. For example, one cannot simply output a single element, as it is impossible to distinguish elements of a list containing one corrupted and one uncorrupted element. To ensure correctness, algorithms under this setting must output a set to make sure the uncorrupted maximum element is included. We first show that any algorithm must output a set of size at least $\min\{n, 2k + 1\}$ to ensure that the uncorrupted maximum is contained in the output set. Restricted to algorithms whose output size is exactly $\min\{n, 2k + 1\}$, for deterministic algorithms, we show matching upper and lower bounds of $\Theta(nk)$ comparison queries to produce a set of elements that contains the uncorrupted maximum. On the randomized side, we propose a 2-stage algorithm that, with high probability, uses $O(n + k \operatorname{polylog} k)$ comparison queries to find such a set, almost matching the $\Omega(n)$ queries necessary for any randomized algorithm to obtain a constant probability of being correct.

cs.DS↗

arXiv2024

LiveSpeech: Low-Latency Zero-shot Text-to-Speech via Autoregressive Modeling of Audio Discrete Codes

Prior works have demonstrated zero-shot text-to-speech by using a generative language model on audio tokens obtained via a neural audio codec. It is still challenging, however, to adapt them to low-latency scenarios. In this paper, we present LiveSpeech - a fully autoregressive language model-based approach for zero-shot text-to-speech, enabling low-latency streaming of the output audio. To allow multiple token prediction within a single decoding step, we propose (1) using adaptive codebook loss weights that consider codebook contribution in each frame and focus on hard instances, and (2) grouping codebooks and processing groups in parallel. Experiments show our proposed models achieve competitive results to state-of-the-art baselines in terms of content accuracy, speaker similarity, audio quality, and inference speed while being suitable for low-latency streaming applications.

cs.SD↗

arXiv2024

SiNGR: Brain Tumor Segmentation via Signed Normalized Geodesic Transform Regression

One of the primary challenges in brain tumor segmentation arises from the uncertainty of voxels close to tumor boundaries. However, the conventional process of generating ground truth segmentation masks fails to treat such uncertainties properly. Those "hard labels" with 0s and 1s conceptually influenced the majority of prior studies on brain image segmentation. As a result, tumor segmentation is often solved through voxel classification. In this work, we instead view this problem as a voxel-level regression, where the ground truth represents a certainty mapping from any pixel to the border of the tumor. We propose a novel ground truth label transformation, which is based on a signed geodesic transform, to capture the uncertainty in brain tumors' vicinity. We combine this idea with a Focal-like regression L1-loss that enables effective regression learning in high-dimensional output space by appropriately weighting voxels according to their difficulty. We thoroughly conduct an experimental evaluation to validate the components of our proposed method, compare it to a diverse array of state-of-the-art segmentation models, and show that it is architecture-agnostic. The code of our method is made publicly available (\url{https://github.com/Oulu-IMEDS/SiNGR/}).

cs.CV↗

arXiv2024

Image-level Regression for Uncertainty-aware Retinal Image Segmentation

Accurate retinal vessel (RV) segmentation is a crucial step in the quantitative assessment of retinal vasculature, which is needed for the early detection of retinal diseases and other conditions. Numerous studies have been conducted to tackle the problem of segmenting vessels automatically using a pixel-wise classification approach. The common practice of creating ground truth labels is to categorize pixels as foreground and background. This approach is, however, biased, and it ignores the uncertainty of a human annotator when it comes to annotating e.g. thin vessels. In this work, we propose a simple and effective method that casts the RV segmentation task as an image-level regression. For this purpose, we first introduce a novel Segmentation Annotation Uncertainty-Aware (SAUNA) transform, which adds pixel uncertainty to the ground truth using the pixel's closeness to the annotation boundary and vessel thickness. To train our model with soft labels, we generalize the earlier proposed Jaccard metric loss to arbitrary hypercubes for soft Jaccard index (Intersection-over-Union) optimization. Additionally, we employ a stable version of the Focal-L1 loss for pixel-wise regression. We conduct thorough experiments and compare our method to a diverse set of baselines across 5 retinal image datasets. Our empirical results indicate that the integration of the SAUNA transform and these segmentation losses led to significant performance boosts for different segmentation models. Particularly, our methodology enables UNet-like architectures to substantially outperform computational-intensive baselines. Our implementation is available at \url{https://github.com/Oulu-IMEDS/SAUNA}.

cs.CV↗

arXiv2024

A Multi-Dimensional Online Contention Resolution Scheme for Revenue Maximization

We study multi-buyer multi-item sequential item pricing mechanisms for revenue maximization with the goal of approximating a natural fractional relaxation -- the ex ante optimal revenue. We assume that buyers' values are subadditive but make no assumptions on the value distributions. While the optimal revenue, and therefore also the ex ante benchmark, is inapproximable by any simple mechanism in this context, previous work has shown that a weaker benchmark that optimizes over so-called ``buy-many" mechanisms can be approximable. Approximations are known, in particular, for settings with either a single buyer or many unit-demand buyers. We extend these results to the much broader setting of many subadditive buyers. We show that the ex ante buy-many revenue can be approximated via sequential item pricings to within an $O(\log^2 m)$ factor, where $m$ is the number of items. We also show that a logarithmic dependence on $m$ is necessary. Our approximation is achieved through the construction of a new multi-dimensional Online Contention Resolution Scheme (OCRS), that provides an online rounding of the optimal ex ante solution. Chawla et al. arXiv:2204.01962 previously constructed an OCRS for revenue for unit-demand buyers, but their construction relied heavily on the ``almost single dimensional" nature of unit-demand values. Prior to that work, OCRSes have only been studied in the context of social welfare maximization for single-parameter buyers. For the welfare objective, constant-factor approximations have been demonstrated for a wide range of combinatorial constraints on item allocations and classes of buyer valuation functions. Our work opens up the possibility of a similar success story for revenue maximization.

cs.GT↗

arXiv2024

uaMix-MAE: Efficient Tuning of Pretrained Audio Transformers with Unsupervised Audio Mixtures

Masked Autoencoders (MAEs) learn rich low-level representations from unlabeled data but require substantial labeled data to effectively adapt to downstream tasks. Conversely, Instance Discrimination (ID) emphasizes high-level semantics, offering a potential solution to alleviate annotation requirements in MAEs. Although combining these two approaches can address downstream tasks with limited labeled data, naively integrating ID into MAEs leads to extended training times and high computational costs. To address this challenge, we introduce uaMix-MAE, an efficient ID tuning strategy that leverages unsupervised audio mixtures. Utilizing contrastive tuning, uaMix-MAE aligns the representations of pretrained MAEs, thereby facilitating effective adaptation to task-specific semantics. To optimize the model with small amounts of unlabeled data, we propose an audio mixing technique that manipulates audio samples in both input and virtual label spaces. Experiments in low/few-shot settings demonstrate that \modelname achieves 4-6% accuracy improvements over various benchmarks when tuned with limited unlabeled data, such as AudioSet-20K. Code is available at https://github.com/PLAN-Lab/uamix-MAE

cs.SD↗

arXiv2023

Optimality in Mean Estimation: Beyond Worst-Case, Beyond Sub-Gaussian, and Beyond $1+\alpha$ Moments

There is growing interest in improving our algorithmic understanding of fundamental statistical problems such as mean estimation, driven by the goal of understanding the limits of what we can extract from valuable data. The state of the art results for mean estimation in $\mathbb{R}$ are 1) the optimal sub-Gaussian mean estimator by [LV22], with the tight sub-Gaussian constant for all distributions with finite but unknown variance, and 2) the analysis of the median-of-means algorithm by [BCL13] and a lower bound by [DLLO16], characterizing the big-O optimal errors for distributions for which only a $1+\alpha$ moment exists for $\alpha \in (0,1)$. Both results, however, are optimal only in the worst case. We initiate the fine-grained study of the mean estimation problem: Can algorithms leverage useful features of the input distribution to beat the sub-Gaussian rate, without explicit knowledge of such features? We resolve this question with an unexpectedly nuanced answer: "Yes in limited regimes, but in general no". For any distribution $p$ with a finite mean, we construct a distribution $q$ whose mean is well-separated from $p$'s, yet $p$ and $q$ are not distinguishable with high probability, and $q$ further preserves $p$'s moments up to constants. The main consequence is that no reasonable estimator can asymptotically achieve better than the sub-Gaussian error rate for any distribution, matching the worst-case result of [LV22]. More generally, we introduce a new definitional framework to analyze the fine-grained optimality of algorithms, which we call "neighborhood optimality", interpolating between the unattainably strong "instance optimality" and the trivially weak "admissibility" definitions. Applying the new framework, we show that median-of-means is neighborhood optimal, up to constant factors. It is open to find a neighborhood-optimal estimator without constant factor slackness.

math.ST↗

arXiv2023

Improving Pearson's chi-squared test: hypothesis testing of distributions -- optimally

Pearson's chi-squared test, from 1900, is the standard statistical tool for "hypothesis testing on distributions": namely, given samples from an unknown distribution $Q$ that may or may not equal a hypothesis distribution $P$, we want to return "yes" if $P=Q$ and "no" if $P$ is far from $Q$. While the chi-squared test is easy to use, it has been known for a while that it is not "data efficient", it does not make the best use of its data. Precisely, for accuracy $\epsilon$ and confidence $\delta$, and given $n$ samples from the unknown distribution $Q$, a tester should return "yes" with probability $>1-\delta$ when $P=Q$, and "no" with probability $>1-\delta$ when $|P-Q|>\epsilon$. The challenge is to find a tester with the \emph{best} tradeoff between $\epsilon$, $\delta$, and $n$. We introduce a new tester, efficiently computable and easy to use, which we hope will replace the chi-squared tester in practical use. Our tester is found via a new non-convex optimization framework that essentially seeks to "find the tester whose Chernoff bounds on its performance are as good as possible". This tester is $1+o(1)$ optimal, in that the number of samples $n$ needed by the tester is within $1+o(1)$ factor of the samples needed by \emph{any} tester, even non-linear testers (for the setting: accuracy $\epsilon$, confidence $\delta$, and hypothesis $P$). We complement this algorithmic framework with matching lower bounds saying, essentially, that "our tester is instance-optimal, even to $1+o(1)$ factors, to the degree that Chernoff bounds are tight". Our overall non-convex optimization framework extends well beyond the current problem and is of independent interest.

math.ST↗

arXiv2023

ConsistencyTTA: Accelerating Diffusion-Based Text-to-Audio Generation with Consistency Distillation

Diffusion models are instrumental in text-to-audio (TTA) generation. Unfortunately, they suffer from slow inference due to an excessive number of queries to the underlying denoising network per generation. To address this bottleneck, we introduce ConsistencyTTA, a framework requiring only a single non-autoregressive network query, thereby accelerating TTA by hundreds of times. We achieve so by proposing "CFG-aware latent consistency model," which adapts consistency generation into a latent space and incorporates classifier-free guidance (CFG) into model training. Moreover, unlike diffusion models, ConsistencyTTA can be finetuned closed-loop with audio-space text-aware metrics, such as CLAP score, to further enhance the generations. Our objective and subjective evaluation on the AudioCaps dataset shows that compared to diffusion-based counterparts, ConsistencyTTA reduces inference computation by 400x while retaining generation quality and diversity.

cs.SD↗

arXiv2023

Reward Selection with Noisy Observations

We study a fundamental problem in optimization under uncertainty. There are $n$ boxes; each box $i$ contains a hidden reward $x_i$. Rewards are drawn i.i.d. from an unknown distribution $\mathcal{D}$. For each box $i$, we see $y_i$, an unbiased estimate of its reward, which is drawn from a Normal distribution with known standard deviation $\sigma_i$ (and an unknown mean $x_i$). Our task is to select a single box, with the goal of maximizing our reward. This problem captures a wide range of applications, e.g. ad auctions, where the hidden reward is the click-through rate of an ad. Previous work in this model [BKMR12] proves that the naive policy, which selects the box with the largest estimate $y_i$, is suboptimal, and suggests a linear policy, which selects the box $i$ with the largest $y_i - c \cdot \sigma_i$, for some $c > 0$. However, no formal guarantees are given about the performance of either policy (e.g., whether their expected reward is within some factor of the optimal policy's reward). In this work, we prove that both the naive policy and the linear policy are arbitrarily bad compared to the optimal policy, even when $\mathcal{D}$ is well-behaved, e.g. has monotone hazard rate (MHR), and even under a "small tail" condition, which requires that not too many boxes have arbitrarily large noise. On the flip side, we propose a simple threshold policy that gives a constant approximation to the reward of a prophet (who knows the realized values $x_1, \dots, x_n$) under the same "small tail" condition. We prove that when this condition is not satisfied, even an optimal clairvoyant policy (that knows $\mathcal{D}$) cannot get a constant approximation to the prophet, even for MHR distributions, implying that our threshold policy is optimal against the prophet benchmark, up to constants.

cs.GT↗

arXiv2022

A Multi-scale Graph Signature for Persistence Diagrams based on Return Probabilities of Random Walks

Persistence diagrams (PDs), often characterized as sets of death and birth of homology class, have been known for providing a topological representation of a graph structure, which is often useful in machine learning tasks. Prior works rely on a single graph signature to construct PDs. In this paper, we explore the use of a family of multi-scale graph signatures to enhance the robustness of topological features. We propose a deep learning architecture to handle this set input. Experiments on benchmark graph classification datasets demonstrate that our proposed architecture outperforms other persistent homology-based methods and achieves competitive performance compared to state-of-the-art methods using graph neural networks. In addition, our approach can be easily applied to large size of input graphs as it does not suffer from limited scalability which can be an issue for graph kernel methods.

cs.LG↗

arXiv2022

A Study on Self-Supervised Object Detection Pretraining

In this work, we study different approaches to self-supervised pretraining of object detection models. We first design a general framework to learn a spatially consistent dense representation from an image, by randomly sampling and projecting boxes to each augmented view and maximizing the similarity between corresponding box features. We study existing design choices in the literature, such as box generation, feature extraction strategies, and using multiple views inspired by its success on instance-level image representation learning techniques. Our results suggest that the method is robust to different choices of hyperparameters, and using multiple views is not as effective as shown for instance-level image representation learning. We also design two auxiliary tasks to predict boxes in one view from their features in the other view, by (1) predicting boxes from the sampled set by using a contrastive loss, and (2) predicting box coordinates using a transformer, which potentially benefits downstream object detection tasks. We found that these tasks do not lead to better object detection performance when finetuning the pretrained model on labeled data.

cs.CV↗