While the Machine Learning (ML) landscape is evolving rapidly, there has been a relative lag in the development of the "learning systems" needed to enable broad adoption. Furthermore, few such systems are designed to support the specialized requirements of scientific ML. Here we present the Data and Learning Hub for sc…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,657 papers · 148 categories
Trend · papers per month
48 results for “model serving”
HOLMES improves real-time model serving for ICU patients, balancing accuracy and speed.
problem Real-time model serving is crucial in ICU due to urgency and cost.
method Online model ensemble serving framework for healthcare applications.
result HOLMES achieves high accuracy and sub-second latency for model ensemble serving.
Machine Learning models are often composed of pipelines of transformations. While this design allows to efficiently execute single model components at training time, prediction serving has different requirements such as low latency, high throughput and graceful performance degradation under heavy load. Current predicti…
New principle reduces load imbalance in LLM serving systems, saving up to 52% energy.
problem Wasted computational power due to load imbalance in LLM serving systems.
method Developed a universal load-balancing principle for barrier-synchronized systems with non-migratable state.
result Proves worst-case theoretical guarantees for imbalance reduction and energy savings.
DeepLight: Deep Lightweight Feature Interactions for Accelerating CTR Predictions in Ad Servingcs.LG
Click-through rate (CTR) prediction is a crucial task in online display advertising. The embedding-based neural networks have been proposed to learn both explicit feature interactions through a shallow component and deep feature interactions using a deep neural network (DNN) component. These sophisticated models, howev…
We introduce a generative adversarial network (GAN) model to simulate the 3-dimensional Lagrangian motion of particles trapped in the recirculation zone of a buoyancy-opposed flame. The GAN model comprises a stochastic recurrent neural network, serving as a generator, and a convoluted neural network, serving as a discr…
New algorithm poLinUCB improves online learning in content recommendation platforms.
problem Improving efficiency in content recommendation platforms with post-serving context.
method Novel contextual bandit problem with post-serving contexts and a new algorithm, poLinUCB.
result Achieves tight regret under standard assumptions and significant benefit of utilizing post-serving contexts.
Service-induced congestion in memory-constrained LLM serving
problem Service-induced congestion in memory-constrained large language model (LLM) serving
method Developing a discrete-time dynamical model of memory-constrained LLM inference
result The system converges to a unique worst-case limit cycle that is asymptotically stable outside a Lebesgue-measure-zero exact-capture set, with throughput losses as large as 50%.
The integration of artificial intelligence capabilities into modern software systems is increasingly being simplified through the use of cloud-based machine learning services and representational state transfer architecture design. However, insufficient information regarding underlying model provenance and the lack of …
While current machine learning models have impressive performance over a wide range of applications, their large size and complexity render them unsuitable for tasks such as remote monitoring on edge devices with limited storage and computational power. A naive approach to resolve this on the model level is to use simp…
Paper proposes a method to aggregate customer engagement data for better ranking of e-commerce results.
problem Cold start problem and under-representation of new or under-impressed products in e-commerce search results.
method Aggregates customer engagements within a day for the same query as input training data for machine learning models.
result Training models on aggregated data leads to better ranking of new and under-impressed products.
A new invariant for braids and knots uses secants and trisecants.
problem Creating an invariant for braids and knots.
method Defining secant-quandle (SQ) with secants and trisecants as generators and relations.
result Secant-quandle (SQ) is a novel invariant for braids and knots.
We construct algebraic and algebro-geometric models for the spaces of unparametrized paths. This is done by considering a path as a holonomy functional on indeterminate connections. For a manifold X, we construct a Lie algebroid P which serves as the tangent space to X (punctual paths) inside the space of all unparamet…
Proposes a flexible neural recommendation framework for better prediction performance.
problem Data sparsity, cold start problem, and long-tail distribution in recommendations.
method A modular neural recommendation framework that includes a neural collaborative filtering part and a text processing part as a regularizer.
result Achieves better prediction performance than state-of-the-art text-aware methods using a simple text processing approach.
DRIFT uses neural flows to replace distributional regression models.
problem Lack of neural network representations for distributional regression models.
method Inverse flow transformations (DRIFT) for distributional regression.
result Neural representations in DRIFT match classical statistical methods in performance.
We introduce Microsoft Machine Learning for Apache Spark (MMLSpark), an ecosystem of enhancements that expand the Apache Spark distributed computing library to tackle problems in Deep Learning, Micro-Service Orchestration, Gradient Boosting, Model Interpretability, and other areas of modern computation. Furthermore, we…
Financial models are studied where each asset may potentially lose value relative to any other. Conditioning on non-devaluation, each asset can serve as proper numéraire and classical valuation rules can be formulated. It is shown when and how these local valuation rules can be aggregated to obtain global arbitrage-fre…
New priors improve robustness and interpretability in penalized regression.
problem Improper priors in penalized regression lead to suboptimal solutions.
method Developed non-zero priors inspired by human decision heuristics.
result Robust priors yield excellent worst-case performance across various tasks.
We introduce Dimple, a fully open-source API for probabilistic modeling. Dimple allows the user to specify probabilistic models in the form of graphical models, Bayesian networks, or factor graphs, and performs inference (by automatically deriving an inference engine from a variety of algorithms) on the model. Dimple a…
Improved language models using ratio-matching and KL divergence.
problem Efficiently modeling discrete data with diffusion models.
method Introduced new theorems and a novel CTMC transition-rate matrix for ratio-matching and KL divergence.
result 10-15% improvement in perplexity and faster training steps.
Deep neural networks are vulnerable to adversarial examples, which poses security concerns on these algorithms due to the potentially severe consequences. Adversarial attacks serve as an important surrogate to evaluate the robustness of deep learning models before they are deployed. However, most of existing adversaria…
Linear recurrent networks explain reinforcement learning performance in partially observable settings.
problem Understanding why linear recurrent networks work in reinforcement learning with partial observability.
method Constructed and studied two linear filters for HMMs and action-controlled HMMs.
result Linear filters serve as sufficient statistics and reduce state ambiguity, explaining empirical reinforcement learning success.
Random forests remain among the most popular off-the-shelf supervised machine learning tools with a well-established track record of predictive accuracy in both regression and classification settings. Despite their empirical success as well as a bevy of recent work investigating their statistical properties, a full and…
This paper is devoted to the problem of choosing the most suitable model of a geometrical system for describing the real crystallographic space. It has been shown that all 230 crystallographic groups used to describe the crystalline structures in a Euclidean space can be presented by elliptic motions in the closed spac…
The Predictive Normalized Maximum Likelihood (pNML) scheme has been recently suggested for universal learning in the individual setting, where both the training and test samples are individual data. The goal of universal learning is to compete with a ``genie'' or reference learner that knows the data values, but is res…
Proposes a method to retrain neural networks incrementally for continuous data flow.
problem Continuous data flow and the challenges of catastrophic forgetting and efficient retraining.
method Incremental retraining using multi-armed bandits to select important samples and weights, and a new regularization term for synapse and neuron importance.
result Mitigates catastrophic forgetting and boosts model performance.
We describe some of the algebra underlying the decomposition of planar grid diagrams. This provides a useful toy model for an extension of Heegaard Floer homology to 3-manifolds with parametrized boundary. This paper is meant to serve as a gentle introduction to the subject, and does not itself have immediate topologic…
Minimal Variance Hedging of Options with Student-t Underlyingcond-mat.stat-mech
I explicitly work out closed form solutions for the optimal hedging strategies (in the sense of Bouchaud and Sornette) in the case of European call options, where the underlying is modeled by (unbiased) iid additive returns with Student-t distributions. The results may serve as illustrative examples for option pricing …
We extend Kirman's model by introducing variable event time scale. The proposed flexible time scale is equivalent to the variable trading activity observed in financial markets. Stochastic version of the extended Kirman's agent based model is compared to the non-linear stochastic models of long-range memory in financia…
We present a novel microscopic stock market model consisting of a large number of random agents modeling traders in a market. Each agent is characterized by a set of parameters that serve to make iterated predictions of two successive returns. The future price is determined according to the offer and the demand of all …
Empirical analysis serves as an important complement to theoretical analysis for studying practical Bayesian optimization. Often empirical insights expose strengths and weaknesses inaccessible to theoretical analysis. We define two metrics for comparing the performance of Bayesian optimization methods and propose a ran…
Access to food assistance programs such as food pantries and food banks needs focus in order to mitigate food insecurity. Accessibility to the food assistance programs is impacted by demographics of the population and geography of the location. It hence becomes imperative to define and identify food assistance deserts …
A Bernstein-type inequality for stochastic processes of quadratic forms of Gaussian variablesmath.ST
We introduce a Bernstein-type inequality which serves to uniformly control quadratic forms of gaussian variables. The latter can for example be used to derive sharp model selection criteria for linear estimation in linear regression and linear inverse problems via penalization, and we do not exclude that its scope of a…
The paper introduces uncertainty quantification for NER models.
problem Current NER models lack uncertainty measures, leading to downstream errors.
method Full-Sequence and Subsequence Conformal Prediction framework.
result The method provides formal guarantees about the reliability of model predictions.
Gradient estimate for linearized translator equation in R^4.
problem Analyzing singularity models of mean curvature flow in R^4.
method Proving a gradient estimate for the variation field W in the tip region.
result Sharp bound for the derivative of the variation field W in the tip region.
Aiming at a comprehensive and concise tutorial survey, recap of variational inference and reinforcement learning with Probabilistic Graphical Models are given with detailed derivations. Reviews and comparisons on recent advances in deep reinforcement learning are made from various aspects. We offer detailed derivations…
Towards Regulatory-Confirmed Adaptive Clinical Trials: Machine Learning Opportunities and Solutionsstat.ML
New framework for adaptive clinical trials to address real-world challenges.
problem Real-world challenges in post-regulatory clinical trials.
method RFAN framework integrating regulatory constraints and treatment policy value.
result Empirical evaluation of RFAN's performance.
We investigate the problem of modeling symbolic sequences of polyphonic music in a completely general piano-roll representation. We introduce a probabilistic model based on distribution estimators conditioned on a recurrent neural network that is able to discover temporal dependencies in high-dimensional sequences. Our…
The availability of large amounts of time series data, paired with the performance of deep-learning algorithms on a broad class of problems, has recently led to significant interest in the use of sequence-to-sequence models for time series forecasting. We provide the first theoretical analysis of this time series forec…
This book was intended to serve as supporting material for a mini-course on web geometry delivered at the 27th Brazilian Mathematical Colloquium which took place at IMPA in the last week of July 2009.
Tennis is a popular sport worldwide, boasting millions of fans and numerous national and international tournaments. Like many sports, tennis has benefitted from the popularity of rigorous record-keeping of game and player information, as well as the growth of machine learning methods for use in sports analytics. Of par…
This article serves as errata of the book "Energy of knots and conformal geometry", Series on Knots and Everything Vol. 33, World Scientific, Singapore, 304 pages, (2003). (ver. 27/05/2007)
We solve a broad class of sequential decision-making problems with partially observed states.
problem Sequential decision-making under uncertainty with partially observed states.
method Modeling as a partially observed Markov decision process (POMDP) and separating state and modulation process.
result The approach allows for specialized approximate solution procedures.
Given a metric space of finite asymptotic dimension, we consider a quasi-isometric invariant of the space called dimension function. The space is said to have asymptotic Assouad-Nagata dimension less or equal if there is a linear dimension function in this dimension. We prove that if is a tree-graded space …
Statistical learning theory provides the theoretical basis for many of today's machine learning algorithms. In this article we attempt to give a gentle, non-technical overview over the key ideas and insights of statistical learning theory. We target at a broad audience, not necessarily machine learning researchers. Thi…
FinGPT uses LLMs for real-time market sentiment analysis.
problem Real-time market sentiment analysis for trading.
method Synthesizes financial news and social media data, integrates with technical indicators, uses FinGPT for sentiment analysis.
result Generates actionable trading signals using LLMs.
Characterizes preferences for decision-making under uncertainty using a leader-follower game model.
problem Decision-making under uncertainty and ambiguity aversion.
method Characterizes niveloidal preferences through a leader-follower game model, satisfying specific axioms.
result The leader's strategy space can serve as an ambiguity aversion index.
We explain how the distributional index of a transversally elliptic operator on a principal G-manifold P that is obtained by lifting a Dirac operator on P/G can serve as a link between the Duflo isomorphism and Chern-Weil forms.