Quantum machine learning at LHCb: First proton-proton collisions at a world-record energy announced

With its brand-new detec­tor designed to han­dle sig­nif­i­cant­ly more chal­leng­ing data-tak­ing con­di­tions, the Quan­tum machine learn­ing (QML) at the Large Hadron Col­lid­er beau­ty (LHCb) exper­i­ment at Con­seil Européen pour la Recherche Nucléaire (CERN) recent­ly announced the first pro­ton-pro­ton col­li­sions at a world-record ener­gy.

The DPA team, led by Uni­ver­si­ty of Liv­er­pool senior research physi­cist Eduar­do Rodrigues, has demon­strat­ed for the first time the suc­cess­ful use of Quan­tum machine learn­ing tech­niques for the iden­ti­fi­ca­tion of the charge of b‑quark ini­ti­at­ed jets at the The Large Hadron Col­lid­er (LHC).

Quantum machine learning, quantum computing and their ‘interaction’

While machine-learn­ing algo­rithms are used to process enor­mous amounts of data, quan­tum machine learn­ing makes use of qubits, quan­tum oper­a­tions, or spe­cif­ic quan­tum sys­tems to boost the speed and accu­ra­cy of com­pu­ta­tion and data storage.

The quan­tum com­put­er pro­vides a com­plete­ly new type of com­put­ing hard­ware to the machine learn­ing hard­ware pool: quan­tum com­put­ers. The quan­tum the­o­ry explains the com­plete­ly dis­tinct phys­i­cal prin­ci­ples that sup­port infor­ma­tion pro­cess­ing on quan­tum computers.

On the oth­er hand, quan­tum com­put­ing is a way of com­put­ing that relies on the prin­ci­ples of quan­tum mechan­ics to func­tion. Data is encod­ed in bits, which can only be either 1 or 0, in tra­di­tion­al com­put­ing. On the oth­er hand, qubits, which can be both 1 and 0, are used in quan­tum computing.

Quan­tum machine learn­ing inves­ti­gates how con­cepts from quan­tum com­put­ing and machine learn­ing inter­act. The hard­ware we use to run our algo­rithms has always defined the lim­its of what com­put­ers can learn; for instance, par­al­lel graph­ics pro­cess­ing unit (GPU) clus­ters enable the suc­cess of mod­ern deep learn­ing with neur­al networks.

Mod­ern deep learn­ing algo­rithms are applied to large data sets, but the hard­ware lim­its the size of these data sets. To go beyond this lim­it, we need new approach­es that make use of quan­tum capabilities.

Application of quantum machine learning

Quan­tum machine learn­ing is a way for­ward for quan­tum com­put­ing and machine learn­ing research. It com­bines the best of both worlds: Quan­tum com­put­ers can solve cer­tain class­es of prob­lems in far less time than tra­di­tion­al com­put­ers, for exam­ple those relat­ed to opti­miza­tion or machine learn­ing. The the­o­ry behind quan­tum machine learn­ing is still under devel­op­ment, but the first suc­cess­es are being seen in prac­ti­cal applications.

For exam­ple, quan­tum machine learn­ing can help in improv­ing pat­tern recog­ni­tion, which in turn, will make it eas­i­er for sci­en­tists to pre­dict extreme weath­er events and poten­tial­ly save thou­sands of lives a year. Devel­op­ing a room-tem­per­a­ture super­con­duc­tor, elim­i­nat­ing car­bon diox­ide for a bet­ter envi­ron­ment, mak­ing sol­id-state bat­ter­ies, and enhanc­ing the nitro­gen-fix­a­tion process for the pro­duc­tion of ammo­nia-based fer­til­iz­er are some of the urgent prob­lems that could be solved with quan­tum computing.

This paper demon­strat­ed, for the first time, that Quan­tum machine learn­ing can be used with suc­cess in LHCb data analy­sis. — Dr. Eduar­do Rodrigues

More­over, quan­tum machine learn­ing will be cru­cial in devel­op­ing nov­el tech­nolo­gies. Quan­tum machine learn­ing can help us under­stand quan­tum effects and dis­cov­er appli­ca­tions for quan­tum com­put­ing. Anoth­er exam­ple is quan­tum cryp­tog­ra­phy, the abil­i­ty to try to hide infor­ma­tion from eaves­drop­pers, or quan­tum metrol­o­gy, the abil­i­ty to mea­sure and record the dis­tri­b­u­tion of mate­r­i­al prop­er­ties by using a quan­tum computer.

Last but not least, quan­tum machine learn­ing is used for devel­op­ing arti­fi­cial intel­li­gence (AI) algo­rithms that will ana­lyze huge amounts of data in a dif­fer­ent way than clas­si­cal ones.

The com­bi­na­tion of both fields will cer­tain­ly reshape soci­ety. It will change the way we live and think when com­bined with blockchain tech­nol­o­gy, which is also a com­bi­na­tion of dif­fer­ent sci­en­tif­ic fields, such as cryp­tog­ra­phy, dis­trib­uted sys­tems, and peer-to-peer networking.

Related Reading:

Our future depends on the con­ver­gence of tech­nolo­gies and their inte­gra­tion into our every­day lives. But let’s go back to quan­tum machine learn­ing and the lat­est exper­i­ment con­duct­ed by the Data Pro­cess­ing & Analy­sis (DPA) team of researchers.

Experiment by the DPA team

The DPA project is a major ren­o­va­tion of the offline analy­sis frame­work to allow full exploita­tion of the sig­nif­i­cant increase in data flow from the upgrad­ed LHCb detector.

For the medi­um and longer term, this effort is a part of R&D beyond the just-start­ing new data tak­ing period.

In LHCb analy­sis, the usage of machine learn­ing tech­niques is com­mon. Giv­en the quick devel­op­ment of quan­tum com­put­ing and quan­tum tech­nolo­gies, it seems sen­si­ble to start look­ing into whether and how quan­tum algo­rithms can func­tion on this new hard­ware, as well as whether or not the LHCb par­ti­cle physics use-cas­es may prof­it from the grow­ing field of quan­tum computing.

The team used QML tech­niques for the first time to deal with the task of hadron­ic jet charge iden­ti­fi­ca­tion. Till now, QML tech­niques have most­ly been used in par­ti­cle physics to solve event clas­si­fi­ca­tion and par­ti­cle track recon­struc­tion challenges.

Based on a sam­ple of sim­u­lat­ed b‑quark ini­ti­at­ed jets, the study “Quan­tum Machine Learn­ing for b‑jet charge iden­ti­fi­ca­tion” was accom­plished. A Deep Neur­al Net­work (DNN), a mod­ern, pow­er­ful type of con­ven­tion­al (i.e., non-quan­tum) arti­fi­cial intel­li­gence algo­rithm, was used to com­pare the per­for­mance of a so-called Vari­a­tion­al Quan­tum Clas­si­fi­er that is based on two dif­fer­ent quan­tum cir­cuits. Although tests on actu­al hard­ware are now being devel­oped, the per­for­mance is exam­ined using a quan­tum sim­u­la­tor because the quan­tum hard­ware that is cur­rent­ly on the mar­ket is still in its ear­ly stages.

Results

The DNN worked mar­gin­al­ly bet­ter than the quan­tum machine learn­ing algo­rithms when the results were com­pared to those pro­duced using a clas­si­cal Deep Neur­al Network.

The study shows that the quan­tum method of machine learn­ing achieves max­i­mum per­for­mance with min­i­mal events. As a result, it assists in low­er­ing resource uti­liza­tion, which will become vital at LHCb giv­en the vol­ume of data obtained in com­ing years. How­ev­er, the DNN sur­pass­es Quan­tum meth­ods for machine learn­ing when more fea­tures are used. With the avail­abil­i­ty of more effec­tive quan­tum hard­ware, improve­ments are expected.

Quan­tum algo­rithms can be used to study cor­re­la­tions between fea­tures, accord­ing to research con­duct­ed in part­ner­ship with sub­ject mat­ter experts. This would allow it to gath­er data on cor­re­la­tions between jet con­stituents, which would enhance the accu­ra­cy of iden­ti­fy­ing jet fla­vors.

“This paper demon­strat­ed, for the first time, that Quan­tum machine learn­ing can be used with suc­cess in LHCb data analy­sis”, Dr. Eduar­do Rodrigues said. Quan­tum machine learn­ing is just begin­ning to be used in par­ti­cle physics exper­i­ments. Because of the wide­spread inter­est and invest­ment in quan­tum com­put­ing, sig­nif­i­cant advances in hard­ware and com­put­er tech­nolo­gies are to be expect­ed as sci­en­tists devel­op expe­ri­ence with it.

“This work, which is part of the R&D activ­i­ties of the LHCb DPA project, pro­vid­ed valu­able insight into Quan­tum machine learn­ing. The inter­est­ing (first) results open new avenues for clas­si­fi­ca­tion prob­lems in par­ti­cle physics exper­i­ments”, Dr. Rodrigues added.

Future implications of the findings

From the study, it is obvi­ous that the DNN has a bet­ter per­for­mance than the clas­si­cal machine learn­ing algo­rithms when the results were com­pared to those pro­duced using a con­ven­tion­al Deep Neur­al Network.

The study found that sig­nif­i­cant per­for­mance gains can be obtained for the DNN algo­rithms, which could have direct impli­ca­tions on future events analy­sis and recon­struc­tion at LHCb.

In the future, improve­ments to the hard­ware could lead to even more impres­sive per­for­mance. The next step could be to test the per­for­mance of these meth­ods on actu­al data of real b‑quark jets at LHCb.

More­over, this find­ing will enable the imme­di­ate imple­men­ta­tion of Quan­tum machine learn­ing tech­niques into the analy­sis of real LHCb data. Researchers believe that Quan­tum machine learn­ing will sig­nif­i­cant­ly boost the over­all per­for­mance of the LHCb exper­i­ment, pro­vid­ing vital infor­ma­tion for a deep­er under­stand­ing of physics beyond the Stan­dard Mod­el and ulti­mate­ly new discoveries.

Leave a Reply

Your email address will not be published.

Join our NewsletterDaily Glimple of Future

Our blog, "Daily Glimpse of Future", strives to make the future much clearer than it is today. Join our newsletter for free now.