Dynamic Learning Rate for Deep Reinforcement Learning: A Bandit Approach

Autor: Donâncio, Henrique, Barrier, Antoine, South, Leah F., Forbes, Florence

In Deep Reinforcement Learning models trained using gradient-based techniques, the choice of optimizer and its learning rate are crucial to achieving good performance: higher learning rates can prevent the model from learning effectively, while lower

Externí odkaz: http://arxiv.org/abs/2410.12598

Open Problems in Technical AI Governance

Autor: Reuel, Anka, Bucknall, Ben, Casper, Stephen, Fist, Tim, Soder, Lisa, Aarne, Onni, Hammond, Lewis, Ibrahim, Lujain, Chan, Alan, Wills, Peter, Anderljung, Markus, Garfinkel, Ben, Heim, Lennart, Trask, Andrew, Mukobi, Gabriel, Schaeffer, Rylan, Baker, Mauricio, Hooker, Sara, Solaiman, Irene, Luccioni, Alexandra Sasha, Rajkumar, Nitarshan, Moës, Nicolas, Ladish, Jeffrey, Guha, Neel, Newman, Jessica, Bengio, Yoshua, South, Tobin, Pentland, Alex, Koyejo, Sanmi, Kochenderfer, Mykel J., Trager, Robert

AI progress is creating a growing range of risks and opportunities, but it is often unclear how they should be navigated. In many cases, the barriers and uncertainties faced are at least partly technical. Technical AI governance, referring to technic

Externí odkaz: http://arxiv.org/abs/2407.14981

Data Authenticity, Consent, & Provenance for AI are all broken: what will it take to fix them?

Autor: Longpre, Shayne, Mahari, Robert, Obeng-Marnu, Naana, Brannon, William, South, Tobin, Gero, Katy, Pentland, Sandy, Kabbara, Jad

Publikováno v: Proceedings of ICML 2024, in PMLR 235:32711-32725. URL: https://proceedings.mlr.press/v235/longpre24b.html

New capabilities in foundation models are owed in large part to massive, widely-sourced, and under-documented training data collections. Existing practices in data collection have led to challenges in tracing authenticity, verifying consent, preservi

Externí odkaz: http://arxiv.org/abs/2404.12691

Measurement of the 1-jettiness event shape observable in deep-inelastic electron-proton scattering at HERA

Autor: The H1 collaboration, Andreev, V., Arratia, M., Baghdasaryan, A., Baty, A., Begzsuren, K., Bolz, A., Boudry, V., Brandt, G., Britzger, D., Buniatyan, A., Bystritskaya, L., Campbell, A. J., Avila, K. B. Cantun, Cerny, K., Chekelian, V., Chen, Z., Contreras, J. G., Cvach, J., Dainton, J. B., Daum, K., Deshpande, A., Diaconu, C., Drees, A., Eckerlin, G., Egli, S., Elsen, E., Favart, L., Fedotov, A., Feltesse, J., Fleischer, M., Fomenko, A., Gal, C., Gayler, J., Goerlich, L., Gogitidze, N., Gouzevitch, M., Grab, C., Greenshaw, T., Grindhammer, G., Haidt, D., Henderson, R. C. W., Hessler, J., Hladký, J., Hoffmann, D., Horisberger, R., Hreus, T., Huber, F., Jacobs, P. M., Jacquet, M., Janssen, T., Jung, A. W., Katzy, J., Kiesling, C., Klein, M., Kleinwort, C., Klest, H. T., Kluth, S., Kogler, R., Kostka, P., Kretzschmar, J., Krücker, D., Krüger, K., Landon, M. P. J., Lange, W., Laycock, P., Lee, S. H., Levonian, S., Li, W., Lin, J., Lipka, K., List, B., List, J., Lobodzinski, B., Long, O. R., Malinovski, E., Martyn, H. -U., Maxfield, S. J., Mehta, A., Meyer, A. B., Meyer, J., Mikocki, S., Mikuni, V. M., Mondal, M. M., Müller, K., Nachman, B., Naumann, Th., Newman, P. R., Niebuhr, C., Nowak, G., Olsson, J. E., Ozerov, D., Park, S., Pascaud, C., Patel, G. D., Perez, E., Petrukhin, A., Picuric, I., Pitzl, D., Polifka, R., Preins, S., Radescu, V., Raicevic, N., Ravdandorj, T., Reichelt, D., Reimer, P., Rizvi, E., Robmann, P., Roosen, R., Rostovtsev, A., Rotaru, M., Sankey, D. P. C., Sauter, M., Sauvan, E., Schmitt, S., Schmookler, B. A., Schnell, G., Schoeffel, L., Schöning, A., Schumann, S., Sefkow, F., Shushkevich, S., Soloviev, Y., Sopicki, P., South, D., Specka, A., Steder, M., Stella, B., Stöcker, L., Straumann, U., Sun, C., Sykora, T., Thompson, P. D., Acosta, F. Torales, Traynor, D., Tseepeldorj, B., Tu, Z., Tustin, G., Valkárová, A., Vallée, C., Van Mechelen, P., Wegener, D., Wünsch, E., Žáček, J., Zhang, J., Zhang, Z., Žlebčík, R., Zohrabyan, H., Zomer, F.

The H1 Collaboration reports the first measurement of the 1-jettiness event shape observable $\tau_1^b$ in neutral-current deep-inelastic electron-proton scattering (DIS). The observable $\tau_1^b$ is equivalent to a thrust observable defined in the

Externí odkaz: http://arxiv.org/abs/2403.10109

Control Variates for MCMC

Autor: South, Leah, Sutton, Matthew

This chapter describes several control variate methods for improving estimates of expectations from MCMC.
Comment: To appear in the second edition of the Handbook of MCMC. Associated code available at https://github.com/LeahPrice/CVBookChapter/

Externí odkaz: http://arxiv.org/abs/2402.07349