arXiv subjects

Daniel Ting

Publications and source records attributed to Daniel Ting.

At least 19 recordsLinked to original sources

arXiv2026

Outlier Impact: Detection by Consequences

We introduce a novel outlier detection method that utilizes a concrete measure of the statistical impact of an outlier rather than a vague notion of "unusualness". This allows practitioners to only flag points that materially affect the inferences made from the data while also ensuring these points are anomalous. This is of particular interest in experimentation, where outliers affect the validity of the causal inferences. In particular, we consider a potential outlier's impact on the mean and the False Positive Rate (FPR) of a hypothesis test.

stat.ME↗

arXiv2026

Estimating the True Effect Size Distribution with SIMEX

Large-scale online experimentation produces noisy effect estimates, which can overstate gains and complicate decisions about launches and testing policies. We propose a nonparametric method based on SIMulation-EXtrapolation (SIMEX) to estimate the latent distribution of true effects from estimated average treatment effects with known variances. The method evaluates quantiles after adding progressively more simulated measurement noise and extrapolates the resulting inverse cumulative distribution function to the zero-noise setting while enforcing monotonicity of the quantiles. In a synthetic example with normally distributed true effects and measurement error, the method recovers the underlying effect distribution and performs nearly as well as a parametric empirical Bayes normal means approach. This provides a flexible way to characterize effect-size distributions without fully specifying a parametric model.

stat.ME↗

arXiv2026

Position: Explanation Stability Is a Property of the Model Method Pair, Not the Model

This position paper argues that claims about explanation stability are scientifically invalid without cross method validation. Just as statistical significance requires the test statistic to be specified, stability should either be evaluated across multiple attribution paradigms or explicitly scoped to the computational objective of a single method. In controlled chest X ray experiments, DenseNet201, ResNet50V2, and InceptionV3 achieved AUC values above 99%, yet their stability rankings reversed across attribution methods. LayerCAM ranked InceptionV3 as the most stable model, with an IoU of 0.777, whereas GradCAM++ favored DenseNet201 and reduced InceptionV3 stability score by 17.3%. These findings demonstrate that explanation stability is an emergent property of the model method pair rather than an intrinsic characteristic of the model alone. We therefore argue that explanation based claims should be validated across multiple attribution methods and that regulatory submissions should explicitly specify the attribution operators used to avoid creating illusory safety assurances.

cs.CV↗

arXiv2026

Quantifying Explanation Consistency: The C-Score Metric for CAM-Based Explainability in Medical Image Classification

Class Activation Mapping (CAM) methods are widely used to generate visual explanations for deep learning classifiers in medical imaging. However, existing evaluation frameworks assess whether explanations are correct, measured by localisation fidelity against radiologist annotations, rather than whether they are consistent: whether the model applies the same spatial reasoning strategy across different patients with the same pathology. We propose the C-Score (Consistency Score), a confidence-weighted, annotation-free metric that quantifies intra-class explanation reproducibility via intensity-emphasised pairwise soft IoU across correctly classified instances. We evaluate six CAM techniques: GradCAM, GradCAM++, LayerCAM, EigenCAM, ScoreCAM, and MS GradCAM++ across three CNN architectures (DenseNet201, InceptionV3, ResNet50V2) over thirty training epochs on the Kermany chest X-ray dataset, covering transfer learning and fine-tuning phases. We identify three distinct mechanisms of AUC-consistency dissociation, invisible to standard classification metrics: threshold-mediated gold list collapse, technique-specific attribution collapse at peak AUC, and class-level consistency masking in global aggregation. C-Score provides an early warning signal of impending model instability. ScoreCAM deterioration on ResNet50V2 is detectable one full checkpoint before catastrophic AUC collapse and yields architecture-specific clinical deployment recommendations grounded in explanation quality rather than predictive ranking alone.

cs.CV↗

arXiv2026

When Fine-Tuning Changes the Evidence: Architecture-Dependent Semantic Drift in Chest X-Ray Explanations

Transfer learning followed by fine-tuning is widely adopted in medical image classification due to consistent gains in diagnostic performance. However, in multi-class settings with overlapping visual features, improvements in accuracy do not guarantee stability of the visual evidence used to support predictions. We define semantic drift as systematic changes in the attribution structure supporting a model's predictions between transfer learning and full fine-tuning, reflecting potential shifts in underlying visual reasoning despite stable classification performance. Using a five-class chest X-ray task, we evaluate DenseNet201, ResNet50V2, and InceptionV3 under a two-stage training protocol and quantify drift with reference-free metrics capturing spatial localization and structural consistency of attribution maps. Across architectures, coarse anatomical localization remains stable, while overlap IoU reveals pronounced architecture-dependent reorganization of evidential structure. Beyond single-method analysis, stability rankings can reverse across LayerCAM and GradCAM++ under converged predictive performance, establishing explanation stability as an interaction between architecture, optimization phase, and attribution objective.

cs.CV↗

arXiv2025

Evaluating Variance Estimates with Relative Efficiency

Experimentation platforms in industry must often deal with customer trust issues. Platforms must prove the validity of their claims as well as catch issues that arise. As a central quantity estimated by experimentation platforms, the validity of confidence intervals is of particular concern. To ensure confidence intervals are reliable, we must understand and diagnose when our variance estimates are biased or noisy, or when the confidence intervals may be incorrect. A common method for this is A/A testing, in which both the control and test arms receive the same treatment. One can then test if the empirical false positive rate (FPR) deviates substantially from the target FPR over many tests. However, this approach turns each A/A test into a simple binary random variable. It is an inefficient estimate of the FPR as it throws away information about the magnitude of each experiment result. We show how to empirically evaluate the effectiveness of statistics that monitor the variance estimates that partly dictate a platform's statistical reliability. We also show that statistics other than empirical FPR are more effective at detecting issues. In particular, we propose a $t^2$-statistic that is more sample efficient.

stat.ME↗

arXiv2025

The Bias-Variance Tradeoff in Long-Term Experimentation

As we exhaust methods that reduces variance without introducing bias, reducing variance in experiments often requires accepting some bias, using methods like winsorization or surrogate metrics. While this bias-variance tradeoff can be optimized for individual experiments, bias may accumulate over time, raising concerns for long-term optimization. We analyze whether bias is ever acceptable when it can accumulate, and show that a bias-variance tradeoff persists in long-term settings. Improving signal-to-noise remains beneficial, even if it introduces bias. This implies we should shift from thinking there is a single ``correct'', unbiased metric to thinking about how to make the best estimates and decisions when better precision can be achieved at the expense of bias. Furthermore, our model adds nuance to previous findings that suggest less stringent launch criterion leads to improved gains. We show while this is beneficial when the system is far from the optimum, more stringent launch criterion is preferable as the system matures.

stat.ME↗

arXiv2024

Challenges Faced by Large Language Models in Solving Multi-Agent Flocking

Flocking is a behavior where multiple agents in a system attempt to stay close to each other while avoiding collision and maintaining a desired formation. This is observed in the natural world and has applications in robotics, including natural disaster search and rescue, wild animal tracking, and perimeter surveillance and patrol. Recently, large language models (LLMs) have displayed an impressive ability to solve various collaboration tasks as individual decision-makers. Solving multi-agent flocking with LLMs would demonstrate their usefulness in situations requiring spatial and decentralized decision-making. Yet, when LLM-powered agents are tasked with implementing multi-agent flocking, they fall short of the desired behavior. After extensive testing, we find that agents with LLMs as individual decision-makers typically opt to converge on the average of their initial positions or diverge from each other. After breaking the problem down, we discover that LLMs cannot understand maintaining a shape or keeping a distance in a meaningful way. Solving multi-agent flocking with LLMs would enhance their ability to understand collaborative spatial reasoning and lay a foundation for addressing more complex multi-agent tasks. This paper discusses the challenges LLMs face in multi-agent flocking and suggests areas for future improvement and research.

cs.AI↗

arXiv2023

On the Limits of Regression Adjustment

Regression adjustment, sometimes known as Controlled-experiment Using Pre-Experiment Data (CUPED), is an important technique in internet experimentation. It decreases the variance of effect size estimates, often cutting confidence interval widths in half or more while never making them worse. It does so by carefully regressing the goal metric against pre-experiment features to reduce the variance. The tremendous gains of regression adjustment begs the question: How much better can we do by engineering better features from pre-experiment data, for example by using machine learning techniques or synthetic controls? Could we even reduce the variance in our effect sizes arbitrarily close to zero with the right predictors? Unfortunately, our answer is negative. A simple form of regression adjustment, which uses just the pre-experiment values of the goal metric, captures most of the benefit. Specifically, under a mild assumption that observations closer in time are easier to predict that ones further away in time, we upper bound the potential gains of more sophisticated feature engineering, with respect to the gains of this simple form of regression adjustment. The maximum reduction in variance is $50\%$ in Theorem 1, or equivalently, the confidence interval width can be reduced by at most an additional $29\%$.

stat.ME↗

arXiv2023

Sketch-Flip-Merge: Mergeable Sketches for Private Distinct Counting

Data sketching is a critical tool for distinct counting, enabling multisets to be represented by compact summaries that admit fast cardinality estimates. Because sketches may be merged to summarize multiset unions, they are a basic building block in data warehouses. Although many practical sketches for cardinality estimation exist, none provide privacy when merging. We propose the first practical cardinality sketches that are simultaneously mergeable, differentially private (DP), and have low empirical errors. These introduce a novel randomized algorithm for performing logical operations on noisy bits, a tight privacy analysis, and provably optimal estimation. Our sketches dramatically outperform existing theoretical solutions in simulations and on real-world data.

cs.DS↗

arXiv2022

Order-Invariant Cardinality Estimators Are Differentially Private

We consider privacy in the context of streaming algorithms for cardinality estimation. We show that a large class of algorithms all satisfy $\epsilon$-differential privacy, so long as (a) the algorithm is combined with a simple down-sampling procedure, and (b) the cardinality of the input stream is $\Omega(k/\epsilon)$. Here, $k$ is a certain parameter of the sketch that is always at most the sketch size in bits, but is typically much smaller. We also show that, even with no modification, algorithms in our class satisfy $(\epsilon, \delta)$-differential privacy, where $\delta$ falls exponentially with the stream cardinality. Our analysis applies to essentially all popular cardinality estimation algorithms, and substantially generalizes and tightens privacy bounds from earlier works.

cs.DS↗

arXiv2021

Few-Shot Domain Adaptation with Polymorphic Transformers

Deep neural networks (DNNs) trained on one set of medical images often experience severe performance drop on unseen test images, due to various domain discrepancy between the training images (source domain) and the test images (target domain), which raises a domain adaptation issue. In clinical settings, it is difficult to collect enough annotated target domain data in a short period. Few-shot domain adaptation, i.e., adapting a trained model with a handful of annotations, is highly practical and useful in this case. In this paper, we propose a Polymorphic Transformer (Polyformer), which can be incorporated into any DNN backbones for few-shot domain adaptation. Specifically, after the polyformer layer is inserted into a model trained on the source domain, it extracts a set of prototype embeddings, which can be viewed as a "basis" of the source-domain features. On the target domain, the polyformer layer adapts by only updating a projection layer which controls the interactions between image features and the prototype embeddings. All other model weights (except BatchNorm parameters) are frozen during adaptation. Thus, the chance of overfitting the annotations is greatly reduced, and the model can perform robustly on the target domain after being trained on a few annotated images. We demonstrate the effectiveness of Polyformer on two medical segmentation tasks (i.e., optic disc/cup segmentation, and polyp segmentation). The source code of Polyformer is released at https://github.com/askerlee/segtran.

cs.CV↗

arXiv2021

Simple, Optimal Algorithms for Random Sampling Without Replacement

Consider the fundamental problem of drawing a simple random sample of size k without replacement from [n] := {1, . . . , n}. Although a number of classical algorithms exist for this problem, we construct algorithms that are even simpler, easier to implement, and have optimal space and time complexity.

cs.DS↗

arXiv2020

HyperLogLog (HLL) Security: Inflating Cardinality Estimates

Counting the number of distinct elements on a set is needed in many applications, for example to track the number of unique users in Internet services or the number of distinct flows on a network. In many cases, an estimate rather than the exact value is sufficient and thus many algorithms for cardinality estimation that significantly reduce the memory and computation requirements have been proposed. Among them, Hyperloglog has been widely adopted in both software and hardware implementations. The security of Hyperloglog has been recently studied showing that an attacker can create a set of elements that produces a cardinality estimate that is much smaller than the real cardinality of the set. This set can be used for example to evade detection systems that use Hyperloglog. In this paper, the security of Hyperloglog is considered from the opposite angle: the attacker wants to create a small set that when inserted on the Hyperloglog produces a large cardinality estimate. This set can be used to trigger false alarms in detection systems that use Hyperloglog but more interestingly, it can be potentially used to inflate the visits to websites or the number of hits of online advertisements. Our analysis shows that an attacker can create a set with a number of elements equal to the number of registers used in the Hyperloglog implementation that produces any arbitrary cardinality estimate. This has been validated in two commercial implementations of Hyperloglog: Presto and Redis. Based on those results, we also consider the protection of Hyperloglog against such an attack.

cs.CR↗

arXiv2020

Manifold Learning via Manifold Deflation

Nonlinear dimensionality reduction methods provide a valuable means to visualize and interpret high-dimensional data. However, many popular methods can fail dramatically, even on simple two-dimensional manifolds, due to problems such as vulnerability to noise, repeated eigendirections, holes in convex bodies, and boundary bias. We derive an embedding method for Riemannian manifolds that iteratively uses single-coordinate estimates to eliminate dimensions from an underlying differential operator, thus "deflating" it. These differential operators have been shown to characterize any local, spectral dimensionality reduction method. The key to our method is a novel, incremental tangent space estimator that incorporates global structure as coordinates are added. We prove its consistency when the coordinates converge to true coordinates. Empirically, we show our algorithm recovers novel and interesting embeddings on real-world and synthetic datasets.

stat.ML↗

arXiv2020

Conditional Cuckoo Filters

Bloom filters, cuckoo filters, and other approximate set membership sketches have a wide range of applications. Oftentimes, expensive operations can be skipped if an item is not in a data set. These filters provide an inexpensive, memory efficient way to test if an item is in a set and avoid unnecessary operations. Existing sketches only allow membership testing for single set. However, in some applications such as join processing, the relevant set is not fixed and is determined by a set of predicates. We propose the Conditional Cuckoo Filter, a simple modification of the cuckoo filter that allows for set membership testing given predicates on a pre-computed sketch. This filter also introduces a novel chaining technique that enables cuckoo filters to handle insertion of duplicate keys. We evaluate our methods on a join processing application and show that they significantly reduce the number of tuples that a join must process.

cs.DS↗

arXiv2020

Security of HyperLogLog (HLL) Cardinality Estimation: Vulnerabilities and Protection

Count distinct or cardinality estimates are widely used in network monitoring for security. They can be used, for example, to detect the malware spread, network scans, or a denial of service attack. There are many algorithms to estimate cardinality. Among those, HyperLogLog (HLL) has been one of the most widely adopted. HLL is simple, provides good cardinality estimates over a wide range of values, requires a small amount of memory, and allows merging of estimates from different sources. However, as HLL is increasingly used to detect attacks, it can itself become the target of attackers that want to avoid being detected. To the best of our knowledge, the security of HLL has not been studied before. In this letter, we take an initial step in its study by first exposing a vulnerability of HLL that allows an attacker to manipulate its estimate. This shows the importance of designing secure HLL implementations. In the second part of the letter, we propose an efficient protection technique to detect and avoid the HLL manipulation. The results presented strongly suggest that the security of HLL should be further studied given that it is widely adopted in many networking and computing applications.

cs.CR↗

arXiv2018

Count-Min: Optimal Estimation and Tight Error Bounds using Empirical Error Distributions

The Count-Min sketch is an important and well-studied data summarization method. It allows one to estimate the count of any item in a stream using a small, fixed size data sketch. However, the accuracy of the sketch depends on characteristics of the underlying data. This has led to a number of count estimation procedures which work well in one scenario but perform poorly in others. A practitioner is faced with two basic, unanswered questions. Which variant should be chosen when the data is unknown? Given an estimate, is its error sufficiently small to be trustworthy? We provide answers to these questions. We derive new count estimators, including a provably optimal estimator, which best or match previous estimators in all scenarios. We also provide practical, tight error bounds at query time for both new and existing estimators. These error estimates also yield procedures to choose the sketch tuning parameters optimally, as they can extrapolate the error to different choices of sketch width and depth. The key observation is that the distribution of errors in each counter can be empirically estimated from the sketch itself. By first estimating this distribution, count estimation becomes a statistical estimation and inference problem with a known error distribution. This provides both a principled way to derive new and optimal estimators as well as a way to study the error and properties of existing estimators.

cs.DS↗