Memory split advantage: thinner networks outperform a single wide network.
problem Optimizing deep learning models with limited memory.
method Investigated training a single wide network vs. an ensemble of thinner networks with the same total number of parameters.
result An ensemble of several thinner networks outperforms a single wide network for large memory budgets.
This work investigates power laws in deep neural network ensembles and predicts their performance.
problem Understanding the performance of deep neural network ensembles and their optimal structure.
method Investigated the behavior of negative log-likelihood (CNLL) of a deep ensemble as a function of ensemble size and member network size, identifying power law dependencies.
result One large network may perform worse than an ensemble of several medium-size networks, known as a memory split.
Memory bandwidth bottleneck is a major challenges in processing machine learning (ML) algorithms. In-memory acceleration has potential to address this problem; however, it needs to address two challenges. First, in-memory accelerator should be general enough to support a large set of different ML algorithms. Second, it…
Backpropagation-free trunk training improves model performance on various benchmarks.
problem Memory inefficiency and noisy gradient estimates in deep network training.
method Split Forward Gradient (Split-FG) method that splits network into trunk and head, estimating only trunk gradient.
result Split-FG achieves better performance than pure forward-gradient training and backpropagation on various benchmarks.
Split conformal prediction works well for time series despite temporal dependence.
problem Uncertainty quantification for time series predictions with past data.
method Split conformal prediction method for time series data with predictors having memory.
result Theoretical bounds on coverage probability for split conformal prediction in time series with memory.
PSI-LinUCB improves scalability for large recommender systems.
problem Efficiently training and inferring for large action spaces in recommender systems.
method Represent inverse design matrix as diagonal + low-rank correction, derive stable rank-1 and batched updates, use projector-splitting integrator.
result Demonstrated effectiveness on recommender system datasets, achieving scalable training and inference.
Neural network tackles continual learning with neuromodulation and local error signals.
problem Catastrophic forgetting in continuous learning.
method Biologically-inspired neural architecture with local learning and neuromodulation, combined with transfer metalearning.
result Superior performance in continual learning tasks compared to other approaches.
BatchEnsemble reduces ensemble costs by 3X in training and testing.
problem High costs for training and testing ensembles of neural networks.
method Defines each weight matrix as a Hadamard product of a shared matrix and a rank-one matrix per member.
result Achieves 3X speedup and 3X memory reduction in test time for ensembles of size 4.
Study validates Lillo-Mike-Farmer model predicting financial market long-range correlations.
problem Quantifying long-range correlations in financial markets.
method Analyzed nine years of market data to classify traders as order-splitting or random, measured metaorder-length distributions, and compared to LMF model predictions.
result Agreement between LMF model predictions and actual data, validating the model.
Recent empirical studies have demonstrated long-memory in the signs of orders to buy or sell in financial markets [2, 19]. We show how this can be caused by delays in market clearing. Under the common practice of order splitting, large orders are broken up into pieces and executed incrementally. If the size of such lar…
Delay-SDE-net models time series with memory and uncertainty, outperforming other models.
problem Accurately modeling time series with memory and uncertainty.
method Stochastic delay differential equations (SDDEs) neural network model with aleatoric and epistemic uncertainty.
result The Delay-SDE-net consistently outperforms other models in predicting time series values and uncertainties.
Deep learning typically requires training a very capable architecture using large datasets. However, many important learning problems demand an ability to draw valid inferences from small size datasets, and such problems pose a particular challenge for deep learning. In this regard, various researches on "meta-learning…
We introduce the C++ application and R package ranger. The software is a fast implementation of random forests for high dimensional data. Ensembles of classification, regression and survival trees are supported. We describe the implementation, provide examples, validate the package with a reference implementation, and …
A new algorithm splits Gaussian processes for efficient streaming data.
problem Poor scaling of Gaussian processes in streaming data.
method Sequential partitioning of input space and localized Gaussian process fitting.
result The algorithm achieves linear memory complexity and superior time and space complexity.
A new approach for efficient data compression in split DNN computing.
problem Optimizing data compression for DNN models split between mobile devices and edge servers.
method Systematic design and training of bottleneck units that can be inserted at the split point.
result Achieves excellent rate-distortion performance with minimal compute and storage overhead.
Divide-and-conquer method splits large data sets for efficient analysis.
problem Handling large data sets that exceed computational limits.
method Split data into smaller sets, analyze each separately, then combine results.
result Combined results provide statistical inference similar to analyzing entire data set.
Regularizes decision trees to reduce inference time by up to 4x with minimal accuracy loss.
problem Optimizing decision tree execution time on resource-constrained devices.
method Regularizes impurity computation during CART algorithm training to favor highly asymmetric distributions.
result Reduces inference time by up to 4x with minimal accuracy loss.
Along with developing of Peaceman-Rachford Splittling Method (PRSM), many batch algorithms based on it have been studied very deeply. But almost no algorithm focused on the performance of stochastic version of PRSM. In this paper, we propose a new stochastic algorithm based on PRSM, prove its convergence rate in ergodi…
Estimating volatility from recent high frequency data, we revisit the question of the smoothness of the volatility process. Our main result is that log-volatility behaves essentially as a fractional Brownian motion with Hurst exponent H of order 0.1, at any reasonable time scale. This leads us to adopt the fractional s…
Study non-oblivious adversarial bandits with delayed feedback and propose algorithms with improved regret bounds.
problem Adversarial bandit problem with delayed, composite anonymous feedback.
method Propose wrapper algorithm for non-oblivious delay setting, achieving o(T) policy regret. result Achieve o(T) policy regret for many adversarial bandit problems with bounded memory loss sequences. A new model decomposes equity returns and volatilities into memory components.
problem Understanding long-term equity dynamics and volatility patterns.
method Proposes a multivariate generalization of the variance ratio to decompose long-horizon equity dynamics.
result Identifies a five-factor model capturing persistent, antipersistent, and multi-scale memory in returns and volatility.
Batch-splitting (data-parallelism) is the dominant distributed Deep Neural Network (DNN) training strategy, due to its universal applicability and its amenability to Single-Program-Multiple-Data (SPMD) programming. However, batch-splitting suffers from problems including the inability to train very large models (due to…
This paper studies the inference problem in quantile regression (QR) for a large sample size n but under a limited memory constraint, where the memory can only store a small batch of data of size m. A natural method is the naïve divide-and-conquer approach, which splits data into batches of size m, computes the l…
This paper proposes an alternative to E2E training for deep networks, reducing memory footprint.
problem High GPUs memory footprint in end-to-end training of deep networks.
method Locally supervised learning with information propagation loss to avoid information collapse.
result The proposed method achieves competitive performance with less than 40% memory footprint compared to E2E training.
In this paper, we present a novel massively parallel algorithm for accelerating the decision tree building procedure on GPUs (Graphics Processing Units), which is a crucial step in Gradient Boosted Decision Tree (GBDT) and random forests training. Previous GPU based tree building algorithms are based on parallel multi-…
NHC learns scalable algorithmic solutions from diverse tasks.
problem Neural networks struggle to learn algorithmic strategies.
method Memory-augmented network with abstraction mechanism and evolutionary training.
result Reliable learning of robust and scalable algorithmic solutions.
We develop an online learning method for prediction, which is important in problems with large and/or streaming data sets. We formulate the learning approach using a covariance-fitting methodology, and show that the resulting predictor has desirable computational and distribution-free properties: It is implemented onli…
SpaceNet improves continual learning by intelligently compressing neural connections.
problem Catastrophic forgetting in sequential learning tasks.
method SpaceNet trains sparse deep neural networks adaptively, compressing task-specific connections.
result SpaceNet outperforms existing methods in class incremental learning scenarios.
Study evaluates stock price forecasting models during the pandemic.
problem Forecasting stock prices during the Covid-19 pandemic.
method Four models (Long-Short Term Memory, XGBoost, Autoregression, Last Value) were tested on stock prices of Facebook, Amazon, Tesla, Google, and Apple.
result Autoregression and Last Value models outperform other models due to strong correlation between prices.
In big data image/video analytics, we encounter the problem of learning an overcomplete dictionary for sparse representation from a large training dataset, which can not be processed at once because of storage and computational constraints. To tackle the problem of dictionary learning in such scenarios, we propose an a…
Several deep models, esp. the generative, compare the samples from two distributions (e.g. WAE like AutoEncoder models, set-processing deep networks, etc) in their cost functions. Using all these methods one cannot train the model directly taking small size (in extreme -- one element) batches, due to the fact that samp…
SPlit optimizes dataset splitting for better model performance.
problem Improving model performance through optimal dataset splitting.
method Adapting Support Points (SP) algorithm for subsampling and categorical variables in a sequential nearest neighbor approach.
result SPlit significantly improves worst-case testing performance compared to random splitting.
The paper extends keenness concept to bridge splittings and finds conditions for existence.
problem Extending keenness concept to bridge splittings and finding conditions for existence.
method Extending the concept of keenness to bridge splittings and proving existence conditions.
result Existence of strongly keen (g,b)-splitting of a link with distance n for certain integers g, b, and n. Non-split almost complex supermanifolds and non-split Riemannian supermanifolds are studied. The first obstacle for a splitting is parametrized by group orbits on an infinite dimensional vector space. Further it is shown that non-split structures appear in the first case as deformations of a split reduction and in the …
Study flippable Heegaard splittings in Seifert fibered spaces.
problem Identifying flippable Heegaard splittings in Seifert fibered spaces.
method Examining isotopies that interchange Heegaard splitting sides in Seifert fibered spaces.
result Characterized which Heegaard splittings are flippable in Seifert fibered spaces.
Proposes a novel SAM operator for separate item and relational memories.
problem Limited memory interactions in neural networks.
method Introduces a Self-attentive Associative Memory (SAM) operator to separate item and relational memories.
result Achieves competitive results in various tasks, including geometry, graph, reinforcement learning, and question answering.
We study the self-dual Yang-Mills equations in split signature. We give a special solution, called the basic split instanton, and describe the ADHM construction in the split signature. Moreover a split version of t'Hooft ansatz is described.
Paper proposes a novel SVM method for creating survival trees.
problem Creating non-linear survival trees for right-censored data.
method L2-regularized dipole splitting criteria with kernel methods.
result Non-linear splits using polynomial and Gaussian kernels show similar predictive power but often smaller tree sizes.
New methods improve prediction regions for high-dimensional data.
problem Creating effective prediction regions for high-dimensional data.
method CD-split and HPD-split methods that combine split method and data-driven partition.
result CD-split and HPD-split converge to oracle highest predictive density set and satisfy local and asymptotic conditional validity.
In this paper, we propose a dual memory structure for reinforcement learning algorithms with replay memory. The dual memory consists of a main memory that stores various data and a cache memory that manages the data and trains the reinforcement learning agent efficiently. Experimental results show that the dual memory …
New spheres can split a 4D link in ways not possible in 3D.
problem Exploring how splitting spheres behave in 4D space.
method Constructing specific 2-component surface-links in S4. result Found non-isotopic splitting spheres in S4∖Lm,n. This paper studies properties of weak reducing pairs in critical Heegaard splittings.
problem Characterize weak reducing pairs in critical Heegaard splittings.
method Analyze the properties of weak reducing pairs in critical Heegaard splittings.
result Provide a necessary condition for a Heegaard surface to be critical.
We show that if a split link is obtained from a split link L in S3 by 1/n-Dehn surgery along a trivial knot C, then the link L∪C is splittable. That is to say, it is impossible to obtain a split link from a split link via a non-trivial twisting. As its corollary, we completely determine when a trivial li…
Study evaluates when splitting classifiers can improve performance despite disparate treatment.
problem Impact of disparate treatment in classification models.
method Comparison of split classifiers and group-blind classifiers, quantifying performance improvement.
result Proves an equivalent expression for the benefit-of-splitting which can be efficiently computed.
New methods convert complex link presentations to simpler, recognizable forms.
problem Representing split and composite links in a clear, recognizable way.
method Pocket and flip moves to transform link presentations.
result Pocket moves are the only obstruction to representing split links by split plat presentations.
Stable Hadamard Memory improves reinforcement learning by efficiently managing memory.
problem Memory models struggle in partially observable reinforcement learning environments.
method Introduces a novel memory model using the Hadamard product for efficient memory management and updates.
result Significantly outperforms state-of-the-art memory-based methods on challenging benchmarks.
The splitting number of a link is the minimal number of crossing changes between different components required, on any diagram, to convert it to a split link. We introduce new techniques to compute the splitting number, involving covering links and Alexander invariants. As an application, we completely determine the sp…
New proof of Lorentzian splitting theorems using elliptic operators.
problem Proving splitting theorems in Lorentzian geometry.
method Using a negative homogeneity elliptic p-d'Alembert operator to prove theorems.
result Lorentzian splitting theorems are proven in a framework similar to Riemannian geometry.