About me

Hello! Since October 2018 I am a Chargé de Recherche (junior researcher) at INRIA Rennes - Bretagne Atlantique as a member of the Taran team. From January 2018 to September 2018, I held an INRIA postdoctoral position within Cairn (now Taran), working on numerical approximation algorithms for energy-efficient computing with Olivier Sentieys and Tomofumi Yuki. Until December 2017 I was a postdoctoral researcher in the Numerical Analysis group at Oxford University under the supervision of Nick Trefethen, working on rational approximation algorithms inside the Chebfun project. Before that, I obtained my PhD in Computer Science at École Normale Supérieure de Lyon in December 2016 on work regarding algorithmic and implementation aspects of digital filter design, under the supervision of Nicolas Brisebarre and Guillaume Hanrot. My current research interests revolve around approximation theory, numerical computations, computer arithmetic, convex and integer optimization, and acceleration of deep learning computations.

Contact information

CAIRN, IRISA Rennes
Campus de Beaulieu, 263 avenue du Général Leclerc
F-35042 Rennes cedex
France

E-mail: silviu DOT filip AT inria DOT fr

Teaching

• 2023-present: Accelerators for Machine Learning (lectures and practical labs, see here), ISTIC, Univ. Rennes.
• 2019-present: Operating Systems Fundamentals (lecturer), École Normale Supérieure de Rennes
  
• 2017-2018: Numerical Solution of Differential Equations I (tutor), Mathematical Institute, University of Oxford
  
• 2015-2016: Architecture matérielle et logicielle (teaching assistant) in L2 at Univ. Lyon 1
  
• 2015-2016: Projet Intégré (teaching assistant) in M1 at École Normale Supérieure de Lyon
  
• 2014-2015: Calcul Formel (teaching assistant) in M1 at École Normale Supérieure de Lyon
  
• 2014-2015: Compilation (teaching assistant) in M1 at École Normale Supérieure de Lyon
  
• 2013-2014: Algorithmique 2 (teaching assistant) in L3 at École Normale Supérieure de Lyon
  
• 2013-2014: Projet Programmation (teaching assistant) in L3 at École Normale Supérieure de Lyon
  

Thesis

• Silviu-Ioan Filip. Robust tools for weighted Chebyshev approximation and applications to digital filter design, PhD dissertation, École Normale Supérieure de Lyon, 2016. Available here.

Publications

International peer-reviewed journals

• Martin Kumm, Anastasia Volkova, Silviu-Ioan Filip, Design of Optimal Multiplierless FIR Filters With Minimal Number of Adders, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 42, 2 (Feb. 2023), 658-671, preliminary version available here.
• Silviu-Ioan Filip, Aurya Javeed, Nick Trefethen, Smooth random functions, random ODEs, and Gaussian processes, SIAM Review, 61(1): 185-205, 2019 preliminary version available here.
• Silviu-Ioan Filip, Yuji Nakatsukasa, Nick Trefethen, Bernhard Beckermann, Rational minimax approximation via adaptive barycentric representations, SIAM Journal on Scientific Computing 40(4): A2427-A2455, 2018, preliminary version available here. Supplementary MATLAB code available here.
• Nicolas Brisebarre, Silviu-Ioan Filip, Guillaume Hanrot. A lattice basis reduction approach for the design of finite wordlength FIR filters, IEEE Transactions on Signal Processing 66, 10 (May 2018), 2673 - 2684, preliminary version available here.
• Silviu-Ioan Filip. A robust and scalable implementation of the Parks-McClellan algorithm for designing FIR filters, ACM Transactions on Mathematical Software 43, 1 (Aug. 2016), 7:1 - 7:24, preliminary version available here.

Publications in peer-reviewed proceedings of international conferences

• Cédric Gernigon, Silviu-Ioan Filip, Olivier Sentieys, Clément Coggiola, Mickaël Bruno. AdaQAT: Adaptive Bit-Width Quantization-Aware Training, to appear in proceeding of the 6th IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), Apr 2024, Abu Dhabi, UAE.
• Léo Pradels, Silviu-Ioan Filip, Olivier Sentieys, Daniel Chillet, Thibaut Le Calloch. FPGA-based CNN Acceleration using Pattern-Aware Pruning, to appear in proceeding of the 6th IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS), Apr 2024, Abu Dhabi, UAE.
• Sami Ben Ali, Silviu-Ioan Filip, Olivier Sentieys. A Stochastic Rounding-Enabled Low-Precision Floating-Point MAC for DNN Training. to appear in proceedings of the 2024 IEEE/ACM Design, Automation and Test in Europe (DATE) conference, Mar 2024, Valencia, Spain. Preliminary version available here.
• Cédric Gernigon, Silviu-Ioan Filip, Olivier Sentieys, Clément Coggiola, Mickael Bruno. Low-Precision Floating-Point for Efficient On-Board Deep Neural Network Processing, proceedings of the 2023 European Data Handling & Data Processing (EDHPC), Oct 2023, Juan-Les-Pins, France. Preliminary version available here.
• Nicolas Brisebarre, Silviu-Ioan Filip. Towards Machine-Efficient Rational \(L^\infty\)-Approximations of Mathematical Functions, proceedings of the 30th IEEE Symposium on Computer Arithmetic (ARITH-30), Portland (Oregon, USA), September 2023. Best paper award winner at the conference. Preliminary version available here.
• Mariko Tatsumi, Silviu-Ioan Filip, Caroline White, Olivier Sentieys, Guy Lemieux. Mixing Low-Precision Formats in Multiply-Accumulate Units for DNN Training, proceedings of the 2022 IEEE International Conference on Field Programmable Technology (ICFPT), pages 1-9, Hong Kong (SAR China), December 2022. Preliminary version available here.
• Olivier Sentieys, Silviu-Ioan Filip, David Briand, David Novo, Etienne Dupuis, Ian O'Connor, Alberto Bosio. AdequateDL: Approximating Deep Learning Accelerators, proceedings of the 24th International Symposium on Design and Diagnostics of Electronic Circuits & Systems (DDECS), pages 37-40, Vienna (Austria), April 2021. Preliminary version available here.
• Florent de Dinechin, Silviu-Ioan Filip, Martin Kumm, Anastasia Volkova. Towards Arithmetic-Centered Filter Design, proceedings of the 28th IEEE Symposium on Computer Arithmetic (ARITH-28), pages 115-118, Torino (Italy), June 2021. Preliminary version available here.
• Florent de Dinechin, Silviu-Ioan Filip, Martin Kumm, Luc Forget. Table-Based versus Shift-And-Add Constant Multiplier for FPGAs, proceedings of the 26th IEEE Symposium on Computer Arithmetic (ARITH-26), pages 151-158, Kyoto (Japan), June 2019. Preliminary version available here.
• Nicolas Brisebarre, George Constantinides, Miloš Ercegovac, Silviu-Ioan Filip, Matei Istoan, Jean-Michel Muller. A High Throughput Polynomial and Rational Function Approximations Evaluator, proceedings of the 25th IEEE Symposium on Computer Arithmetic (ARITH-25), pages 99-106, Amherst (MA, USA), June 2018. Preliminary version available here.

Books

• Etienne Dupuis, Silviu-Ioan Filip, David Novo, Ian O'Connor, Alberto Bosio. Approximations in Deep Learning, in Alberto Bosio, Daniel Ménard, Olivier Sentieys, editors, Approximate Computing Techniques, chapter 15, pages 467-512, Springer, 2022. Preliminary version available here.

Publications in international workshops

• Mariko Tatsumi, Yuxiang Xie, Caroline White, Silviu-Ioan Filip, Olivier Sentieys, Guy Lemieux. MPTorch and MPArchimedes: Open Source Frameworks to Explore Custom Mixed-Precision Operations for DNN Training on Edge Devices, ROAD4NN 2021 - 2nd ROAD4NN Workshop: Research Open Automatic Design for Neural Networks, December 2021. Preliminary version available here.

Publications in the peer-reviewed proceedings of national conferences

• Nicolas Brisebarre, Silviu-Ioan Filip, Guillaume Hanrot. De nouveaux résultats sur la synthèse de filtres RIF, short communication at GRETSI 2015.

Articles under submission, research reports

• Nicolas Brisebarre, Florent De Dinechin, Silviu-Ioan Filip, Matei Istoan. Automatic generation of hardware FIR filters from a frequency domain specification, 2016, research report available here.

PhD Student Supervision

I am lucky to work / have worked with the following students:
• Cédric Gernigon: 2020-present
• Léo Pradels: 2020-present
• Sami Ben Ali: 2022-present

Software

See my Github page. I am also an active contributor to the Chebfun project.

• firpm: A scalable C++ implementation of the Parks-McClellan (Remez) algorithm for designing FIR filters
• fquantizer: Euclidean lattice-based quantization routines written in C++ for digital filter design
• emethodHW: An automatic tool for the evaluation of polynomial and rational function approximations on FPGAs
• FIRopt: A tool for the simultaneous design of digital FIR filters along with the dedicated hardware model
• rminimax: A tool for designing flexible polynomial and rational approximations to mathematical functions (for use inside mathematical libraries)
• MPTorch: A low/mixed-precision simulation wrapper framework built atop PyTorch