| 1. |
|
|
| 2. |
- Träuble, F., et al.
(författare)
-
The Role Of Pretrained Representations For The Ood Generalization Of Rl Agents
- 2022
-
Ingår i: ICLR 2022 - 10th International Conference on Learning Representations. - : International Conference on Learning Representations, ICLR.
-
Konferensbidrag (refereegranskat)abstract
- Building sample-efficient agents that generalize out-of-distribution (OOD) in real-world settings remains a fundamental unsolved problem on the path towards achieving higher-level cognition. One particularly promising approach is to begin with low-dimensional, pretrained representations of our world, which should facilitate efficient downstream learning and generalization. By training 240 representations and over 10,000 reinforcement learning (RL) policies on a simulated robotic setup, we evaluate to what extent different properties of pretrained VAE-based representations affect the OOD generalization of downstream agents. We observe that many agents are surprisingly robust to realistic distribution shifts, including the challenging sim-to-real case. In addition, we find that the generalization performance of a simple downstream proxy task reliably predicts the generalization performance of our RL agents under a wide range of OOD settings. Such proxy tasks can thus be used to select pretrained representations that will lead to agents that generalize.
|
|
| 3. |
- Hamaus, N., et al.
(författare)
-
Euclid : Forecasts from redshift-space distortions and the Alcock-Paczynski test with cosmic voids
- 2022
-
Ingår i: Astronomy and Astrophysics. - : EDP Sciences. - 0004-6361 .- 1432-0746. ; 658
-
Tidskriftsartikel (refereegranskat)abstract
- Euclid is poised to survey galaxies across a cosmological volume of unprecedented size, providing observations of more than a billion objects distributed over a third of the full sky. Approximately 20 million of these galaxies will have their spectroscopy available, allowing us to map the three-dimensional large-scale structure of the Universe in great detail. This paper investigates prospects for the detection of cosmic voids therein and the unique benefit they provide for cosmological studies. In particular, we study the imprints of dynamic (redshift-space) and geometric (Alcock-Paczynski) distortions of average void shapes and their constraining power on the growth of structure and cosmological distance ratios. To this end, we made use of the Flagship mock catalog, a state-of-the-art simulation of the data expected to be observed with Euclid. We arranged the data into four adjacent redshift bins, each of which contains about 11000 voids and we estimated the stacked void-galaxy cross-correlation function in every bin. Fitting a linear-theory model to the data, we obtained constraints on f/b and DMH, where f is the linear growth rate of density fluctuations, b the galaxy bias, D-M the comoving angular diameter distance, and H the Hubble rate. In addition, we marginalized over two nuisance parameters included in our model to account for unknown systematic effects in the analysis. With this approach, Euclid will be able to reach a relative precision of about 4% on measurements of f/b and 0.5% on DMH in each redshift bin. Better modeling or calibration of the nuisance parameters may further increase this precision to 1% and 0.4%, respectively. Our results show that the exploitation of cosmic voids in Euclid will provide competitive constraints on cosmology even as a stand-alone probe. For example, the equation-of-state parameter, w, for dark energy will be measured with a precision of about 10%, consistent with previous more approximate forecasts.
|
|
| 4. |
- Kieran, Mark W., et al.
(författare)
-
A global approach to long-term follow-up of targeted and immune-based therapy in childhood and adolescence
- 2021
-
Ingår i: Pediatric Blood and Cancer. - : Wiley. - 1545-5009 .- 1545-5017. ; 68:7
-
Tidskriftsartikel (refereegranskat)abstract
- While considerable efforts and progress in our understanding of the long-term toxicities of surgery, radiation and chemotherapy in children with cancer have been made over the last 5 decades, there continues to be a wide gap in our knowledge of the long-term health impact of most novel targeted and immunotherapy agents. To address this gap, ACCELERATE, a multi-stakeholder collaboration of clinical and translational academics, regulators from the EMA and FDA, patient/family advocates and members spanning small biotechnology through to large pharmaceutical companies have initiated the development of an international long-term follow-up data registry to collect this important information prospectively. Providing critical safety data on the long-term use of these approved and investigational therapies in children will support the regulatory requirements and labeling information. It will also provide the necessary insight to help guide physicians and families on the appropriateness of a targeted or immune therapy for their child and inform survivorship planning.
|
|
| 5. |
|
|
