Skip to content

Commit 354bda7

Browse files
committed
deploy: f2668ec
1 parent 127c8d6 commit 354bda7

1 file changed

Lines changed: 190 additions & 0 deletions

File tree

references/index.html

Lines changed: 190 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6067,6 +6067,196 @@ <h2 id="year2025" class="year">2025 <span class="count-stat"></span></h2>
60676067
</div>
60686068

60696069

6070+
</div>
6071+
</li>
6072+
<li><div class="bibtex-entry-container">
6073+
<div>
6074+
<div class="col-md-3 col-sm-12 bibtex-ref-meta hidden">
6075+
<div class="row">
6076+
<div class="col-md-12 ref-label label label-default">
6077+
GonzalezEtAl2025
6078+
</div>
6079+
</div>
6080+
</div>
6081+
6082+
<div class="bibtex-ref-entry">
6083+
<span id="GonzalezEtAl2025">X. Gonzalez, L. Kozachkov, D. M. Zoltowski, K. L. Clarkson, and S. W. Linderman, “Predictability Enables Parallelization of Nonlinear State Space Models,” arXiv:2508.16817v4 [math.OC], 2025 [Online]. Available at: <a href="https://arxiv.org/abs/2508.16817v4" target="_blank">https://arxiv.org/abs/2508.16817v4</a></span>
6084+
</div>
6085+
6086+
6087+
<div class="row mt-4">
6088+
<div class="btn-group btn-group-xs btn-group-justified" role="group">
6089+
<div class="btn-group btn-group-xs" role="group">
6090+
6091+
<button class="button" data-toggle="modal" data-target="#modalGonzalezEtAl2025Bibtex" aria-expanded="false"
6092+
aria-controls="modalGonzalezEtAl2025Bibtex">
6093+
BibTeX
6094+
</button>
6095+
6096+
</div>
6097+
<div class="btn-group btn-group-xs" role="group">
6098+
6099+
<button class="button" data-toggle="modal" data-target="#modalGonzalezEtAl2025Abstract" aria-expanded="false"
6100+
aria-controls="modalGonzalezEtAl2025Abstract">
6101+
Abstract
6102+
</button>
6103+
6104+
</div>
6105+
</div>
6106+
</div>
6107+
</div>
6108+
6109+
<div class="modal" id="modalGonzalezEtAl2025Bibtex" tabindex="-1" role="dialog"
6110+
aria-labelledby="modalGonzalezEtAl2025BibtexTitle" aria-hidden="true">
6111+
<div class="modal-background">
6112+
<div class="modal-card">
6113+
<div class="modal-content">
6114+
<header class="modal-card-head">
6115+
<div class="modal-card-head-content">
6116+
<div>
6117+
<p class="modal-card-title" id="modalGonzalezEtAl2025BibtexTitle">
6118+
BibTeX entry <code>GonzalezEtAl2025</code>
6119+
</p>
6120+
</div>
6121+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
6122+
aria-hidden="true">&times;</span></button>
6123+
</div>
6124+
</header>
6125+
<section class="modal-card-body">
6126+
<pre class="abstract">@unpublished{GonzalezEtAl2025,
6127+
author = {Gonzalez, Xavier and Kozachkov, Leo and Zoltowski, David M. and Clarkson, Kenneth L. and Linderman, Scott W.},
6128+
howpublished = {arXiv:2508.16817v4 [math.OC]},
6129+
title = {Predictability Enables Parallelization of Nonlinear State Space Models},
6130+
url = {https://arxiv.org/abs/2508.16817v4},
6131+
year = {2025}
6132+
}
6133+
</pre>
6134+
</section>
6135+
</div>
6136+
</div>
6137+
</div>
6138+
</div>
6139+
6140+
6141+
<div class="modal" id="modalGonzalezEtAl2025Abstract" tabindex="-1" role="dialog"
6142+
aria-labelledby="modalGonzalezEtAl2025AbstractTitle" aria-hidden="true">
6143+
<div class="modal-background">
6144+
<div class="modal-card">
6145+
<div class="modal-content">
6146+
<header class="modal-card-head">
6147+
<div class="modal-card-head-content">
6148+
<div>
6149+
<p class="modal-card-title" id="modalGonzalezEtAl2025AbstractTitle">
6150+
Abstract for BibTeX entry <code>GonzalezEtAl2025</code>
6151+
</p>
6152+
</div>
6153+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
6154+
aria-hidden="true">&times;</span></button>
6155+
</div>
6156+
</header>
6157+
<section class="modal-card-body">
6158+
The rise of parallel computing hardware has made it increasingly important to understand which nonlinear state space models can be efficiently parallelized. Recent advances like DEER (arXiv:2309.12252) and DeepPCR (arXiv:2309.16318) recast sequential evaluation as a parallelizable optimization problem, sometimes yielding dramatic speedups. However, the factors governing the difficulty of these optimization problems remained unclear, limiting broader adoption. In this work, we establish a precise relationship between a system’s dynamics and the conditioning of its corresponding optimization problem, as measured by its Polyak-Lojasiewicz (PL) constant. We show that the predictability of a system, defined as the degree to which small perturbations in state influence future behavior and quantified by the largest Lyapunov exponent (LLE), impacts the number of optimization steps required for evaluation. For predictable systems, the state trajectory can be computed in at worst O((\log T)^2) time, where T is the sequence length: a major improvement over the conventional sequential approach. In contrast, chaotic or unpredictable systems exhibit poor conditioning, with the consequence that parallel evaluation converges too slowly to be useful. Importantly, our theoretical analysis shows that predictable systems always yield well-conditioned optimization problems, whereas unpredictable systems lead to severe conditioning degradation. We validate our claims through extensive experiments, providing practical guidance on when nonlinear dynamical systems can be efficiently parallelized. We highlight predictability as a key design principle for parallelizable models.
6159+
</section>
6160+
</div>
6161+
</div>
6162+
</div>
6163+
</div>
6164+
6165+
</div>
6166+
</li>
6167+
<li><div class="bibtex-entry-container">
6168+
<div>
6169+
<div class="col-md-3 col-sm-12 bibtex-ref-meta hidden">
6170+
<div class="row">
6171+
<div class="col-md-12 ref-label label label-default">
6172+
GonzalezEtAl2025b
6173+
</div>
6174+
</div>
6175+
</div>
6176+
6177+
<div class="bibtex-ref-entry">
6178+
<span id="GonzalezEtAl2025b">X. Gonzalez <i>et al.</i>, “A Unifying Framework for Parallelizing Sequential Models with Linear Dynamical Systems,” arXiv:2509.21716v2 [cs.LG], 2025 [Online]. Available at: <a href="https://arxiv.org/abs/2509.21716v2" target="_blank">https://arxiv.org/abs/2509.21716v2</a></span>
6179+
</div>
6180+
6181+
6182+
<div class="row mt-4">
6183+
<div class="btn-group btn-group-xs btn-group-justified" role="group">
6184+
<div class="btn-group btn-group-xs" role="group">
6185+
6186+
<button class="button" data-toggle="modal" data-target="#modalGonzalezEtAl2025bBibtex" aria-expanded="false"
6187+
aria-controls="modalGonzalezEtAl2025bBibtex">
6188+
BibTeX
6189+
</button>
6190+
6191+
</div>
6192+
<div class="btn-group btn-group-xs" role="group">
6193+
6194+
<button class="button" data-toggle="modal" data-target="#modalGonzalezEtAl2025bAbstract" aria-expanded="false"
6195+
aria-controls="modalGonzalezEtAl2025bAbstract">
6196+
Abstract
6197+
</button>
6198+
6199+
</div>
6200+
</div>
6201+
</div>
6202+
</div>
6203+
6204+
<div class="modal" id="modalGonzalezEtAl2025bBibtex" tabindex="-1" role="dialog"
6205+
aria-labelledby="modalGonzalezEtAl2025bBibtexTitle" aria-hidden="true">
6206+
<div class="modal-background">
6207+
<div class="modal-card">
6208+
<div class="modal-content">
6209+
<header class="modal-card-head">
6210+
<div class="modal-card-head-content">
6211+
<div>
6212+
<p class="modal-card-title" id="modalGonzalezEtAl2025bBibtexTitle">
6213+
BibTeX entry <code>GonzalezEtAl2025b</code>
6214+
</p>
6215+
</div>
6216+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
6217+
aria-hidden="true">&times;</span></button>
6218+
</div>
6219+
</header>
6220+
<section class="modal-card-body">
6221+
<pre class="abstract">@unpublished{GonzalezEtAl2025b,
6222+
author = {Gonzalez, Xavier and Buchanan, E. Kelly and Lee, Hyun Dong and Liu, Jerry Weihong and Wang, Ke Alexander and Zoltowski, David M. and Kozachkov, Leo and Ré, Christopher and Linderman, Scott W.},
6223+
howpublished = {arXiv:2509.21716v2 [cs.LG]},
6224+
title = {A Unifying Framework for Parallelizing Sequential Models with Linear Dynamical Systems},
6225+
url = {https://arxiv.org/abs/2509.21716v2},
6226+
year = {2025}
6227+
}
6228+
</pre>
6229+
</section>
6230+
</div>
6231+
</div>
6232+
</div>
6233+
</div>
6234+
6235+
6236+
<div class="modal" id="modalGonzalezEtAl2025bAbstract" tabindex="-1" role="dialog"
6237+
aria-labelledby="modalGonzalezEtAl2025bAbstractTitle" aria-hidden="true">
6238+
<div class="modal-background">
6239+
<div class="modal-card">
6240+
<div class="modal-content">
6241+
<header class="modal-card-head">
6242+
<div class="modal-card-head-content">
6243+
<div>
6244+
<p class="modal-card-title" id="modalGonzalezEtAl2025bAbstractTitle">
6245+
Abstract for BibTeX entry <code>GonzalezEtAl2025b</code>
6246+
</p>
6247+
</div>
6248+
<button type="button" class="delete" data-dismiss="modal" aria-label="Close"><span
6249+
aria-hidden="true">&times;</span></button>
6250+
</div>
6251+
</header>
6252+
<section class="modal-card-body">
6253+
Harnessing parallelism in seemingly sequential models is a central challenge for modern machine learning. Several approaches have been proposed for evaluating sequential processes in parallel using iterative fixed-point methods, like Newton, Picard, and Jacobi iterations. In this work, we show that these methods can be understood within a common framework based on linear dynamical systems (LDSs), where different iteration schemes arise naturally as approximate linearizations of a nonlinear recursion. Moreover, we theoretically analyze the rates of convergence of these methods, and we verify the predictions of this theory with several case studies. This unifying framework highlights shared principles behind these techniques and clarifies when particular fixed-point methods are most likely to be effective. By bridging diverse algorithms through the language of LDSs, the framework provides a clearer theoretical foundation for parallelizing sequential models and points toward new opportunities for efficient and scalable computation.
6254+
</section>
6255+
</div>
6256+
</div>
6257+
</div>
6258+
</div>
6259+
60706260
</div>
60716261
</li>
60726262
<li><div class="bibtex-entry-container">

0 commit comments

Comments
 (0)