Release notes and future developments

NOMAD 4 is a complete redesign compared with NOMAD 3, with a new architecture providing more flexible code, some added functionalities and reusable code.

Some functionalities available in NOMAD 3 will be included in NOMAD 4 in future releases:

• RobustMads [AudIhaLedTrib2016] and StoMads [G-2019-30]

• DiscoMads [Audet2022a]

• Categorical [AuDe01a] and periodical variables [AuLe2012]

The performance of NOMAD 4 and 3 are similar when the default parameters of NOMAD 3 are used (see [AuLeRoTr2022]).

References

Audet2022a

3. Audet, A. Batailly, S. Kojtych. 2022. Escaping unknown discontinuous regions in blackbox optimization. SIAM Journal on Optimization 32, 3 (2022), 1843-1870.

AudIhaLedTrib2016

3. Audet, A. Ihaddadene, S. Le Digabel, and C. Tribes. 2018. Robust optimization of noisy blackbox problems using the Mesh Adaptive Direct Search algorithm. Optimization Letters 12, 4 (2018), 675–689

G-2019-30

3. Audet, K.J. Dzahini, M. Kokkolaras, and S. Le Digabel. 2021. Stochastic mesh adaptive direct search for blackbox optimization using probabilistic estimates. Technical Report G-2019-30. Les cahiers du GERAD. To appear in Computational Optimization and Applications.

AuDe01a

3. Audet and J.E. Dennis, Jr. 2001. Pattern Search Algorithms for Mixed Variable Programming. SIAM Journal on Optimization 11, 3 (2001), 573–594.

AuLe2012

3. Audet and S. Le Digabel. 2012. The mesh adaptive direct search algorithm for periodic variables. Pacific Journal of Optimization 8, 1 (2012),103–119