The assignee ranking covers 100 companies measured by patent family/record counts, led by a single filer with 3,068 records, well ahead of a fifth-place holder at 594 and a tenth-place holder at 451. The top 10 combined account for 27.2% of the 38,627 records in scope, which means the remaining assignees in the ranking, plus a large population outside it, hold the majority of filings as a long tail. Large diversified electronics and software companies dominate the upper ranks, reflecting reinforcement learning's role as a component technology inside broader AI and systems patents rather than a standalone niche.

Yes, through the last complete filing year in this dataset. Filings rose from 2,294 in 2021 to 3,700 in 2024, a 61% increase over that span, and 2024 is the most recent year that can be treated as a full year of data. 2025 and 2026 counts appear lower only because publication typically lags actual filing by about 18 months, so those years are still filling in rather than reflecting a genuine slowdown.

Core AI computing methods (IPC class G06N) appear in 21.6% of the 38,627 records in scope, and general electric digital data processing (G06F) appears in 16.4%. Beyond that core, image and video processing, wireless communication networks, business data processing and healthcare informatics each carry between roughly 4.5% and 6.1% of records, showing reinforcement learning claims extending well beyond pure algorithm patents into applied domains. Because a single record can carry multiple IPC classes, these shares add up to more than 100% and should not be summed.

Relative to the dense core AI classes, healthcare informatics and business-process applications of reinforcement learning show comparatively low filing density, as do integrations with wireless network resource management and edge inference. These are areas where the underlying reward-based learning methods are documented but their application to specific downstream tasks -- like dynamic model evaluation in resource-constrained network settings -- is less densely claimed. A first claim in these areas would likely combine a known policy-learning or reward-based method with a domain-specific data pipeline or deployment constraint not yet heavily covered.

With the top 10 of 100 ranked assignees holding 27.2% of the 38,627 records in scope, and a leader nearly seven times the size of the tenth-place holder, the field shows a concentrated core sitting above a long tail. Co-assignment is also rare -- only 10 co-assignee pairs were identified across the entire dataset -- indicating that most large portfolios were built through internal filing rather than joint development. This pattern is common in component technologies that get embedded into many companies' broader AI and systems patents rather than filed as a standalone specialty.