High-throughput sequencing technologies generate vast amounts of genomic data in FASTQ format, making efficient compression essential. However, existing methods often prioritize compression performance while neglecting functional features such as two-dimensional random access to individual bases and quality scores. This limitation obstructs large-scale genomic analyses that require selective retrieval of specific data, forcing researchers to resort to costly full-file decompression. To overcome these challenges, we propose FIBRA, an innovative lossless and reference-free compression algorithm that integrates Fibonacci encoding with wavelet trees and employs pruning strategies to significantly reduce storage overhead. Furthermore, FIBRA incorporates a succinct random access index, enabling O(1)-time two-dimensional random access without compromising compression performance. Additionally, the self-synchronizing nature of Fibonacci encoding ensures robust fault tolerance in high-noise sequencing environments, while the inherently parallel architecture delivers excellent performance on both third-generation long-read data and resource-constrained embedded systems. Extensive experiments demonstrate FIBRA's superiority. On dataset SRR035464, it achieves a 36.49% higher compression ratio than Gzip with dramatically faster compression speeds, which is 24.34 times faster than LFQC on SRR004364, and maintains random access latency below 10 ms per access.

Generative Adversarial Imitation Learning (GAIL) stands as a cornerstone approach in imitation learning. This paper investigates the gradient explosion in two types of GAIL: GAIL with deterministic policy (DE-GAIL) and GAIL with stochastic policy (ST-GAIL). We begin with the observation that the training can be highly unstable for DE-GAIL at the beginning of the training phase and end up divergence. Conversely, the ST-GAIL training trajectory remains consistent, reliably converging. To shed light on these disparities, we provide an explanation from a theoretical perspective. By establishing a probabilistic lower bound for GAIL, we demonstrate that gradient explosion is an inevitable outcome for DE-GAIL due to occasionally large expert-imitator policy disparity, whereas ST-GAIL does not have the issue with it. To substantiate our assertion, we illustrate how modifications in the reward function can mitigate the gradient explosion challenge. Finally, we propose CREDO, a simple yet effective strategy that clips the reward function during the training phase, allowing the GAIL to enjoy high data efficiency and stable trainability.